Uploaded by WONG JIA-EN, GRACE _

essential-mathematics-for-economic-analysis-students-manual-5th-ed compress

advertisement
Student’s Manual
Essential Mathematics for
Economic Analysis
5th edition
Knut Sydsæter
Peter Hammond
Andrés Carvajal
Arne Strøm
For further supporting resources, please visit:
www.mymathlab.com/global
Preface
This Student’s solutions manual accompanies Essential Mathematics for Economic Analysis, 5th
Edition, Pearson, 2016. Its main purpose is to provide more detailed solutions to the problems
SM
marked ⊂
⊃ in the book. The answers provided in this manual should be used in combination with
any shorter answers provided in the main test. There are a few cases where only part of the answer
is set out in detail, because the rest follows the same pattern.
We would appreciate suggestions for improvements from our readers, as well as help in weeding
out inaccuracies and errors.
Coventry, Davis, and Oslo, April 2016
Peter Hammond (P.J.Hammond@warwick.ac.uk)
Andrés Carvajal (acarvaes@icloud.com)
Arne Strøm (arne.strom@econ.uio.no)
Contents
1 Essentials of Logic and Set Theory
Review exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
1
2 Algebra
2.3 Rules of algebra
2.4 Fractions . . . .
2.5 Fractional powers
2.6 Inequalities . . .
2.8 Summation . . .
2.11 Double sums . .
Review exercises . . .
.
.
.
.
.
.
.
2
2
2
3
4
5
6
7
.
.
.
.
.
.
.
8
8
9
9
10
11
12
12
.
.
.
.
.
12
12
13
13
13
15
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3 Solving Equations
3.1 Solving equations . . . . . . . . . . . .
3.2 Equations and their parameters . . . .
3.3 Quadratic equations . . . . . . . . . .
3.4 Nonlinear equations . . . . . . . . . .
3.5 Using implication arrows . . . . . . . .
3.6 Two linear equations in two unknowns
Review exercises . . . . . . . . . . . . . . .
4 Functions of One Variable
4.2 Basic definitions . . . .
4.4 Linear functions . . . .
4.6 Quadratic functions . .
4.7 Polynomials . . . . . . .
4.8 Power functions . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
i
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4.10 Logarithmic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Review exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
5 Properties of Functions
5.3 Inverse functions . . .
5.4 Graphs of equations .
5.5 Distance in the plane .
5.6 General functions . . .
Review exercises . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
17
17
17
18
18
18
6 Differentiation
6.2 Tangents and derivatives . . .
6.5 A dash of limits . . . . . . . .
6.7 Sums, products and quotients
6.8 The chain rule . . . . . . . .
6.10 Exponential functions . . . .
6.11 Logarithmic functions . . . .
Review exercises . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
18
18
19
20
21
21
21
23
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
and Newton’s method
. . . . . . . . . . . . .
. . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
23
23
23
24
24
24
24
25
25
25
26
27
27
.
.
.
.
.
.
.
29
29
29
29
30
30
32
32
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
7 Derivatives in Use
7.1 Implicit differentiation . . . . .
7.2 Economic examples . . . . . . .
7.3 Differentiating the inverse . . .
7.4 Linear approximations . . . . .
7.5 Polynomial approximations . .
7.6 Taylor’s formula . . . . . . . .
7.7 Elasticities . . . . . . . . . . . .
7.8 Continuity . . . . . . . . . . . .
7.9 More on limits . . . . . . . . .
7.10 The intermediate value theorem
7.12 L’Hôpital’s rule . . . . . . . . .
Review exercises . . . . . . . . . . .
8 Single-variable Optimization
8.2 Simple tests for extreme points
8.3 Economic examples . . . . . . .
8.4 The extreme value theorem . .
8.5 Further economic examples . .
8.6 Local extreme points . . . . . .
8.7 Inflection points . . . . . . . .
Review exercises . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
9 Integration
34
9.1 Indefinite integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
9.2 Area and definite integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
ii
9.3 Properties of indefinite integrals . . . . . .
9.4 Economic applications . . . . . . . . . . .
9.5 Integration by parts . . . . . . . . . . . .
9.6 Integration by substitution . . . . . . . .
9.7 Infinite intervals of integration . . . . . .
9.8 A glimpse at differential equations . . . .
9.9 Separable and linear differential equations
Review exercises . . . . . . . . . . . . . . . . .
10 Topics in Financial Mathematics
10.2 Continuous compounding . . . .
10.4 Geometric series . . . . . . . . .
10.7 Internal rate of return . . . . . .
Review exercises . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
11 Functions of Many Variables
11.2 Partial derivatives with two variables .
11.3 Geometric representation . . . . . . .
11.5 Functions of more variables . . . . . .
11.6 Partial derivatives with more variables
11.7 Economic applications . . . . . . . . .
11.8 Partial elasticities . . . . . . . . . . . .
Review exercises . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
12 Tools for Comparative Statics
12.1 A simple chain rule . . . . . . . . . . . . .
12.2 Chain rules for many variables . . . . . .
12.3 Implicit differentiation along a level curve
12.4 More general cases . . . . . . . . . . . . .
12.5 Elasticity of substitution . . . . . . . . . .
12.6 Homogeneous functions of two variables .
12.7 Homogeneous and homothetic functions .
12.8 Linear approximations . . . . . . . . . . .
12.9 Differentials . . . . . . . . . . . . . . . . .
12.11Differentiating systems of equations . . .
Review exercises . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
35
36
36
38
39
41
41
42
.
.
.
.
43
43
44
44
45
.
.
.
.
.
.
.
45
45
45
45
46
47
47
47
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
47
47
48
49
49
50
50
51
52
52
53
53
13 Multivariable Optimization
13.2 Two variables: sufficient conditions . . . . . . .
13.3 Local extreme points . . . . . . . . . . . . . . .
13.4 Linear models with quadratic objectives . . . .
13.5 The Extreme Value Theorem . . . . . . . . . .
13.6 The general case . . . . . . . . . . . . . . . . .
13.7 Comparative statics and the Envelope Theorem
Review exercises . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
54
54
55
56
57
60
60
61
iii
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
14 Constrained Optimization
14.1 The Lagrange Multiplier method . . .
14.2 Interpreting the Lagrange Multiplier .
14.3 Multiple solution candidates . . . . . .
14.4 Why the Lagrange method works . . .
14.5 Sufficient conditions . . . . . . . . . .
14.6 Additional variables and constraints .
14.7 Comparative statics . . . . . . . . . .
14.8 Nonlinear programming: a simple case
14.9 Multiple inequality constraints . . . .
14.10Nonnegativity constraints . . . . . . .
Review exercises . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
63
63
64
65
67
68
68
69
70
71
73
74
15 Matrix and Vector Algebra
15.1 Systems of linear equations . . . .
15.3 Matrix multiplication . . . . . . . .
15.4 Rules for matrix multiplication . .
15.5 The transpose . . . . . . . . . . . .
15.6 Gaussian elimination . . . . . . . .
15.8 Geometric interpretation of vectors
15.9 Lines and planes . . . . . . . . . .
Review exercises . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
77
77
77
78
79
79
80
80
81
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
16 Determinants and Inverse Matrices
16.1 Determinants of order 2 . . . . . .
16.2 Determinants of order 3 . . . . . .
16.3 Determinants in general . . . . . .
16.4 Basic rules for determinants . . . .
16.5 Expansion by cofactors . . . . . . .
16.6 The inverse of a matrix . . . . . .
16.7 A general formula for the inverse .
16.8 Cramer’s Rule . . . . . . . . . . . .
Review exercises . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
82
82
82
83
84
84
85
85
86
87
17 Linear Programming
17.1 A graphical approach . . . . . . . .
17.2 Introduction to Duality Theory . .
17.3 The Duality Theorem . . . . . . .
17.4 A general economic interpretation
17.5 Complementary slackness . . . . .
Review exercises . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
89
89
90
91
92
92
93
iv
1 Essentials of Logic and Set Theory
Review exercises for Chapter 1
3. Consider the Venn diagram for three sets depicted in Fig. SM1.R.3. Let nk denote the number
of students in the set marked (k), for k = 1, 2, . . . , 8. Suppose the sets A, B, and C refer to
those who study English, French, and Spanish, respectively. Since 10 students take all three
languages, n7 = 10. There are 15 who take French and Spanish, so 15 = n2 + n7 , and thus
n2 = 5. Furthermore, 32 = n3 + n7 , so n3 = 22. Also, 110 = n1 + n7 , so n1 = 100. The rest of
the information implies that 52 = n2 + n3 + n6 + n7 , so n6 = 52 − 5 − 22 − 10 = 15. Moreover,
220 = n1 + n2 + n5 + n7 , so n5 = 220 − 100 − 5 − 10 = 105. Finally, 780 = n1 + n3 + n4 + n7 ,
so n4 = 780 − 100 − 22 − 10 = 648. The answers are therefore:
(a) n1 = 100,
(b) n3 + n4 = 648 + 22 = 670,
P
(c) 1000 − 7i=1 ni = 1000 − 905 = 95.
F
E
(5)
(1)
(4)
(7)
(2)
(3)
(6)
(8)
S
Figure SM1.R.3
4. (a) ⇒ is true; ⇐ is false, because x = y = 1 also solves x + y = 2.
(b) ⇒ is false, because x2 = 16 also has the solution x = −4; ⇐ true, because if x = 4, then
x2 = 16.
(c) ⇒ is true, because (x − 3)2 ≥ 0; ⇐ false because with y > −2 and x = 3, one has
(x − 3)2 (y + 2) = 0.
(d) ⇒ and ⇐ are both true, since the equation x3 = 8 has the solution x = 2 and no others.1
5. For (a) and (b) see the solutions in the book. For (c), note that when n = 1, the inequality is
obviously correct.2 As the induction hypothesis when n equals the arbitrary natural number k,
suppose that (1 + x)k ≥ 1 + kx. Because 1 + x ≥ 0, we then have
(1 + x)k+1 = (1 + x)k (1 + x) ≥ (1 + kx)(1 + x) = 1 + (k + 1)x + kx2 ≥ 1 + (k + 1)x.
where the last inequality holds because k > 0. Thus, the induction hypothesis holds for
n = k + 1. Therefore, by induction, Bernoulli’s inequality is true for all natural numbers n.
1
2
In the terminology of Section 6.3, function f (x) = x3 is strictly increasing. See Fig. 4.3.7 and Exercise 6.3.3.
And for n = 2, it is correct by part (b).
1
2 Algebra
2.3 Rules of algebra
4. (a) (2t−1)(t2 −2t+1) = 2t(t2 −2t+1)−(t2 −2t+1) = 2t3 −4t2 +2t−t2 +2t−1 = 2t3 −5t2 +4t−1.
(b) (a + 1)2 + (a − 1)2 − 2(a + 1)(a − 1) = (a2 + 2a + 1) + (a2 − 2a + 1) − 2(a2 − 1) = 4.3
(c) (x + y + z)2 = (x + y + z)(x + y + z) = x(x + y + z) + y(x + y + z) + z(x + y + z) =
(x2 + xy + xz) + (yx + y 2 + yz) + (zx + zy + z 2 ) = x2 + y 2 + z 2 + 2xy + 2xz + 2yz.
(d) Put a = x + y + z and b = x − y − z. Then
(x + y + z)2 − (x − y − z)2 = a2 − b2 = (a + b)(a − b) = 2x(2y + 2z) = 4x(y + z)
2.4 Fractions
5. (a)
1
1
x+2
x−2
x+2−x+2
4
−
=
−
=
= 2
.
x−2 x+2
(x − 2)(x + 2) (x + 2)(x − 2)
(x − 2)(x + 2)
x −4
(b) Since 4x + 2 = 2(2x + 1) and 4x2 − 1 = (2x + 1)(2x − 1), the lowest common denominator,
LCD, is 2(2x + 1)(2x − 1). Then,
(6x + 25)(2x − 1) − 2(6x2 + x − 2)
42x − 21
21
6x + 25 6x2 + x − 2
−
=
=
=
.
2
4x + 2
4x − 1
2(2x + 1)(2x − 1)
2(2x + 1)(2x − 1)
2(2x + 1)
a
18b2 − a(a − 3b) + 2(a2 − 9b2 )
a(a + 3b)
a
18b2
−
+
2
=
=
=
.
a2 − 9b2 a + 3b
(a + 3b)(a − 3b)
(a + 3b)(a − 3b)
a − 3b
1
1
(a + 2) − a
2
1
(d)
−
=
=
=
.
8ab 8b(a + 2)
8ab(a + 2)
8ab(a + 2)
4ab(a + 2)
2t − t2
2t
t(2 − t)
3t
−t(t − 2)
3t
−3t2
5t
(e)
·
−
=
·
=
·
=
.
t+2
t−2 t−2
t+2
t−2
t+2
t−2
t+2
1
a 1 − 2a
a− 1
(f) Note that
= 1 2 = 4a − 2, so
0.25
4
1
a 1 − 2a
= 2 − (4a − 2) = 4 − 4a = 4(1 − a).
2−
0.25
(c)
1
2(x + 1) + x − 3x(x + 1)
2 − 3x2
2
+
−3=
=
x x+1
x(x + 1)
x(x + 1)
t
t
t(2t − 1) − t(2t + 1)
−2t
(b)
−
=
= 2
2t + 1 2t − 1
(2t + 1)(2t − 1)
4t − 1
6. (a)
3x
4x
2x − 1
3x(x − 2) + 4x(x + 2) − (2x − 1)
7x2 + 1
−
−
=
= 2
x + 2 2 − x (x − 2)(x + 2)
(x − 2)(x + 2)
x −4
1 1
1 1
+
xy
+
y+x
x y
x y
=
=
(d) The expression equals
= x + y.
1
1
1
· xy
xy
xy
(c)
3
Alternatively, apply the quadratic identity x2 + y 2 − 2xy = (x − y)2 with x = a + 1 and y = a − 1 to obtain
(a + 1)2 + (a − 1)2 − 2(a + 1)(a − 1) = [(a + 1) − (a − 1)]2 = 22 = 4.
2
1
1
1
1
· x2 y 2
−
−
y 2 − x2
x2 y 2
x2 y 2
= 2
.
(e) The expression equals
=
1
1
1
1
y + x2
2y2
+
·
x
+
x2 y 2
x2 y 2
(f) To clear the fractions within both the numerator and denominator, multiply both by xy
to get
a(y − x)
y−x
=
a(y + x)
y+x
1 1
5
4
1
8. (a) − =
−
= , so
4 5
20 20
20
1 1
−
4 5
−2
=
1
20
−2
= 202 = 400.
n2
n·n
n(n − 1) − n2
n
=
=−
.
=n−
1
n
−
1
n
−
1
n
−
1
1− n ·n
1
1
1
1
1
u
(c) Let u = xp−q . Then
+
=
+
=
+
= 1.
p−q
q−p
1+x
1+x
1 + u 1 + 1/u
1+u 1+u
(b) n −
n
1−
1
n
=n−
(d) Using x2 − 1 = (x + 1)(x − 1), one has
1
1
+
(x2 − 1)
(x + 1) + 1
x+2
x − 1 x2 − 1
=
=
,
2 − 1) − 2(x − 1)
2
x(x
(x
−
1)[x(x
+ 1) − 2]
x−
(x2 − 1)
x+1
which reduces to
x+2
1
x+2
=
=
.
2
(x − 1)(x + x − 2)
(x − 1)[(x + 2)(x − 1)]
(x − 1)2
(e) Since
1
1
x2 − (x + h)2
−2xh − h2
−
=
=
,
(x + h)2 x2
x2 (x + h)2
x2 (x + h)2
it follows that
1
1
− 2
2
−2x − h
(x + h)
x
= 2
.
h
x (x + h)2
(f) Multiplying both numerator and denominator by x2 −1 = (x+1)(x−1) yields
which reduces to
2x
.
x−1
10x2
,
5x(x − 1)
2.5 Fractional powers
5. The answers for each respective part that are given in the book emerge after multiplying both
numerator and denominator by the following terms:
√
√
√
√
√
7− 5
(b)
5− 3
(c)
3+2
(a)
√
√
√
√
√
(d) x y − y x
(e)
x+h+ x
(f) 1 − x + 1
2
11. (a) (2x )2 = 22x , which equals 2x if and only if 2x = x2 , or if and only if x = 0 or x = 2.
3
(b) Correct because ap−q = ap /aq .
(c) Correct because a−p = 1/ap .
(d) 51/x = 1/5x = 5−x if and only if 1/x = −x or −x2 = 1, so there is no real x that satisfies
the equation.
(e) Put u = ax and v = ay , which reduces the equation to uv = u + v, or 0 = uv − u − v =
(u − 1)(v − 1) − 1. This is true only for special values of u and v and so for special values
of x and y. In particular, the equation is false when x = y = 1.
√
√
(f) Putting u = x and v = y reduces the equation to 2u · 2v = 2uv , which holds if and
only if uv = u + v, as in (e) above.
2.6 Inequalities
3. (a) This inequality has the same solutions as
3x + 1
3x + 1 − 2(2x + 4)
−x − 7
− 2 > 0, or
> 0, or
> 0.
2x + 4
2x + 4
2x + 4
A sign diagram reveals that the inequality is satisfied for −7 < x < −2. A serious error
is to multiply the inequality by 2x + 4, without checking the sign of 2x + 4. If 2x + 4 < 0,
mulitiplying by this number will reverse the inequality sign.4
(b) The inequality is equivalent to 120/n ≤ 0.75 = 3/4, or (480 − 3n)/4n ≤ 0. A sign
diagram reveals that the inequality is satisfied for n < 0 and for n ≥ 160.5
(c) This is easy: g(g − 2) ≤ 0 and so 0 ≤ g ≤ 2.
p+1
(p − 2) + 3
=
≥ 0.
2
(p − 2)
(p − 2)2
The fraction makes no sense if p = 2. The conclusion that p ≥ −1 and p 6= 2 follows.
(d) Note that p2 − 4p + 4 = (p − 2)2 , so the inequality reduces to
(e) The inequality is equivalent to
−n − 2
−n − 2 − 2n − 8
−3n − 10
10
− 2 > 0 ⇐⇒
> 0 ⇐⇒
> 0 ⇐⇒ −4 < n < −
n+4
n+4
n+4
3
(f) x4 − x2 = x2 (x2 − 1) < 0 ⇐⇒ x 6= 0 and x2 < 1 ⇐⇒ −1 < x < 0 or 0 < x < 1
6. (a) It is easy to see by means of a sign diagram that x(x + 3) < 0 precisely when x lies in
the open interval (−3, 0). Therefore we have ⇒ , but not ⇐: for example, if x = 10,
then x(x + 3) = 130.
(b) x2 < 9 ⇐⇒ −3 < x < 3, so x2 < 9 ⇒ x < 3. If x = −5, for instance, we have x < 3
but x2 > 9, hence we cannot have ⇐ here.
(c) If x > 0, then x2 > 0, but x2 > 0 also when x < 0. So, we have ⇐ but not ⇒.
(d) As y 2 ≥ 0 for any real number y, we have that x > 0 whenever x > y 2 . But x = 1 > 0
does not imply that x > y 2 for y ≥ 1. That is, we have ⇒ but not ⇐ .
4
It could be instructive to test the inequality for some values of x. For example, for x = 0 it is not true. What
about x = −5?
5
Note that for n = 0 the inequality makes no sense. For n = 160, we have equality.
4
√
√
√
9. Note that for any x ≥ 0 and y ≥ 0, x − 2 xy + y = ( x − y)2 ≥ 0, which implies that
1
√
(x + y) ≥ xy.
2
(∗)
Note also that the inequality is strict unless x = y.
To show that mA ≥ mG , simply let x = a and y = b, so that
√
1
mA = (a + b) ≥ ab = mG ,
2
with strict inequality whenever a 6= b.
To show that mG ≥ mH , use x = 1/a and y = 1/b, so that, from (∗),
r
1 1
1 1 1
+
· ≥ 0.
≥
2 a b
a b
Rearranging, we get that
mH
1 1 1 −1
+
≤
=
2 a b
r
1 1
·
a b
!−1
= mG ,
where, again, the inequality is strict unless a = b.
2.8 Summation
3. (a)–(d): In each case, look at the last term in the sum and replace n by k to get an expression
P
for the k th term. Call it sk . Then in (a), (b), and (d) the sum is nk=1 sk , and in (c) we have
Pn
k=0 sk .
(e) The coefficients are the powers 3n for n = 1, 2, 3, 4, 5, so the general term is 3n xn .
(f)–(g) See the answers in the Solutions Section of the book.
(h) This is trickier: one has to see that each term is 198 larger than the previous term.6
Pn
Pn
2
2
2
2
2
2
2
2
7. (a) Valid:
k=1 ck = c · 1 + c · 2 + · · · + c · n = c(1 + 2 + · · · + n ) = c
k=1 k .
(b) Wrong, even for n = 2: the left-hand side is (a1 + a2 )2 = a21 + 2a1 a2 + a22 , but the
right-hand side is a21 + a22 , which is different unless a1 a2 = 0.
(c) Valid: both sides equal b1 + b2 + · · · + bN .
(d) Valid: both sides equal 51 + 52 + 53 + 54 + 55 .
(e) Valid: both sides equal a20,j + · · · + a2n−1,j .
(f) Wrong, even for n = 2: the left-hand side is a1 + a2 /2, but the right-hand side is
(1/k)(a1 + a2 ).
6
This problem is related to the story about Gauss in Section 2.9.
5
2.11 Double sums
1. (a) See the solution in the book.
4 2 X
X
rs 2
rs 2
(b) Note first that
= 0 when s = 0. So
reduces to
r+s
r+s
s=0 r=2
2
X
s=1
"
2s
2+s
2
+
3s
3+s
2
+
4s
4+s
2 #
2 2 2 2 2 2
2
3
4
4
6
8
=
+
+
+
+
+
3
4
5
4
5
6
4
9
16
36 16
(4 + 16) · 400 + 9 · 225 + (16 + 36) · 144
+
+
+1+
+
=
+1
9 16 25
25
9
3600
8000 + 2025 + 7488
17 513
3113
or
+1=
+1=5+
.
3600 P
3600P
P3600
P
P
P
n
n
m
m
n
2
2
(c) Because m
i=1
j=1 (i + j ) =
j=1P i=1 i +
i=1
j=1 j , we can use formulae (2.9.4)
p
and (2.9.5) along with the equality k=1 a = pa to write the sum as
which equals
n
X
1
2 m(m
+ 1) +
m
X
1
6 n(n
+ 1)(2n + 1) = n
i=1
j=1
1
2 m(m
+ 1) + m 61 n(n + 1)(2n + 1)
= 16 mn[3(m + 1) + (n + 1)(2n + 1)] = 16 mn(2n2 + 3n + 3m + 4)
(d) Using formulae (2.9.4) and (2.9.5) again, we get
m X
2
X
ij =
i=1 j=1
m
X
(i + i2 ) =
i=1
=
m
X
i=1
1
6 m(m
i+
m
X
i2 = 21 m(m + 1) + 61 m(m + 1)(2m + 1)
i=1
+ 1)[3 + (2m + 1)] = 16 m(m + 1)(2m + 4) = 13 m(m + 1)(m + 2).
4. ā is the mean of the column means āj , because
n
n
1X
1X
āj =
n
n
j=1
j=1
m
1 X
arj
m
r=1
!
m
=
n
1 XX
arj = ā.
mn
r=1 j=1
To prove (∗), note that because arj − ā is independent of the summation index s, it is a
P
Pm
common factor when we sum over s, so m
s=1 (arj − ā)(asj − ā) = (arj − ā)
s=1 (asj − ā) for
each r. Next, summing over r gives
"m
#" m
#
m X
m
X
X
X
(arj − ā)(asj − ā) =
(arj − ā)
(asj − ā) ,
(∗∗)
r=1 s=1
r=1
s=1
Pm
because s=1 (arj − ā) is a common factor when we sum over r. Using the properties of sums
and the definition of āj , we have
m
X
r=1
(arj − ā) =
m
X
r=1
arj −
m
X
r=1
ā = māj − mā = m(āj − ā).
Similarly, replacing r with s as the index of summation, one also has
m(āj − ā). Substituting these values into (∗∗) then confirms (∗).
6
Pm
s=1 (asj
− ā) =
Review exercises for Chapter 2
5. (a) (2x)4 = 24 x4 = 16x4 .
(b) 2−1 − 4−1 =
1
2
−
1
4
= 41 , so (2−1 − 4−1 )−1 = 4.
(c) Cancel the common factor 4x2 yz 2 .
(d) Here −(−ab3 )−3 = −(−1)−3 a−3 b−9 = a−3 b−9 ,
so [−(−ab3 )−3 (a6 b6 )2 ]3 = [a−3 b−9 a12 b12 ]3 = [a9 b3 ]3 = a27 b9 .
a5 · a3 · a−2
a6
(e)
=
= a3 .
a−3 · a6
a3
−3
3
x
x 3 8 −3
2
= (x5 )−3 = x−15 .
· −2
· 8x
=
(f)
2
x
8
10. Parts (a), (b), (d), (e), and (f) are straightforward; their solutions appear in the book. For
the other parts:
√ √
√ √ √
√ √ √
√
(c) − 3 3 − 6 = −3 + 3 6 = −3 + 3 3 2 = −3 + 3 2.
(g) (1 + x + x2 + x3 )(1 − x) = (1 + x + x2 + x3 ) − (1 + x + x2 + x3 )x = 1 − x4 .
(h) (1 + x)4 = (1 + x)2 (1 + x)2 = (1 + 2x + x2 )(1 + 2x + x2 ), and so on.
12. Parts (a) and (b) are easy, so we focus on the others:
(c) ax + ay + 2x + 2y = a(x + y) + 2(x + y) = (a + 2)(x + y).
(d) 2x2 −5yz+10xz−xy = 2x2 +10xz−(xy+5yz) = 2x(x+5z)−y(x+5z) = (2x−y)(x+5z).
(e) p2 − q 2 + p − q = (p − q)(p + q) + (p − q) = (p − q)(p + q + 1).
(f) u3 + v 3 − u2 v − v 2 u = u2 (u − v) + v 2 (v − u) = (u2 − v 2 )(u − v) = (u + v)(u − v)(u − v),
which simplifies to (u + v)(u − v)2 .
s
s(2s + 1) − s(2s − 1)
2s
s
−
=
= 2
.
2s − 1 2s + 1
(2s − 1)(2s + 1)
4s − 1
x
1−x
24
−x(x + 3) − (1 − x)(x − 3) − 24
−7x − 21
7
(b)
−
− 2
=
=
=
.
3−x x+3 x −9
(x − 3)(x + 3)
(x − 3)(x + 3)
3−x
(c) Multiplying both numerator and denominator by x2 y 2 yields
16. (a)
y−x
y−x
1
=
=
2
2
y −x
(y − x)(y + x)
x+y
17. (a) Simply cancel the factor 25ab.
(b) Factor x2 − y 2 = (x + y)(x − y), and then cancel x + y.
2a − 3b
(2a − 3b)2
=
.
(c) The fraction can be written as
(2a − 3b)(2a + 3b)
2a + 3b
x(4 − x2 )
x(2 − x)(2 + x)
x(2 + x)
4x − x3
=
=
=
(d)
2
2
2
4 − 4x + x
(2 − x)
(2 − x)
2−x
25. Let each side have length s. Then the area K of the equilateral triangle ABC shown in
Fig. SM2.R.25 is the sum of the areas of the three triangles ABP , BCP , and CAP , which
equals 12 sh1 + 21 sh2 + 21 sh3 = K. It follows that h1 + h2 + h3 = 2K/s, which is independent
of where P is placed in the triangle.
7
C
s
h2
h3
s
P
h1
A
B
s
Figure SM2.R.25
29. (a) Using the trick that led to Eq. (2.9.4) in the book,
R=
3+
5+
7 + · · · + 197 + 199 + 201
R = 201 + 199 + 197 + · · · +
7+
5+
3.
Summing vertically term by term, noting that there are 100 terms, gives
2R = 204 + 204 + 204 + · · · + 204 + 204 + 204 = 100 × 204 = 20400,
and thus R = 10200.
(b) Here one has
S = 1001 + 2002 + 3003 + · · · + 8008 + 9009 + 10010
= 1001(1 + 2 + 3 + · · · + 8 + 9 + 10) = 1001 · 55 = 55055
3 Solving Equations
3.1 Solving equations
3. (a) We note first that x = −3 and x = −4 both make the equation absurd. Multiplying the
equation by the common denominator (x+3)(x+4) yields (x−3)(x+4) = (x+3)(x−4),
i.e. x2 + x − 12 = x2 − x − 12, and thus x = 0.
(b) Multiplying by the common denominator (x − 3)(x + 3) yields 3(x + 3) − 2(x − 3) = 9,
from which we get x = −6.
(c) Multiplying by the common denominator 15x, assuming that x 6= 0, yields 18x2 − 75 =
10x2 − 15x + 8x2 , from which we get x = 5.
5. (a) Multiplying by the common denominator 12 yields 9y − 3 − 4 + 4y + 24 = 36y, and so
y = 17/23.
(b) Multiplying by the common denominator 2x(x + 2) yields 8(x + 2) + 6x = 2(2x + 2) + 7x
and so 14x + 16 = 11x + 4, from which we find that 3x = −12 and so x = −4.
(c) Multiplying both numerator and denominator in the first fraction by 1 − z leads to
2 − 2z − z
6
=
.
(1 − z)(1 + z)
2z + 1
Multiplying each side of the equation by (1 − z 2 )(2z + 1) yields (2 − 3z)(2z + 1) = 6 − 6z 2 .
This simplifies to 2 + z − 6z 2 = 6 − 6z 2 whose only solution is z = 4.
8
(d) Expanding all the parentheses gives
p 3 1
p
1 p
1
− − +
− + =− .
4 8 4 12 3 3
3
Multiplying by the common denominator 24 gives the equation 6p − 9 − 6 + 2p − 8 + 8p =
−8, whose solution is p = 15/16.
3.2 Equations and their parameters
2. (a) Multiply both sides by abx to obtain b + a = 2abx. Hence,
b
a
b+a
1 1
1
.
=
+
=2
+
x=
2ab
2ab 2ab
a b
(b) Multiply the equation by cx + d to obtain ax + b = cAx + dA, or (a − cA)x = dA − b,
and thus x = (dA − b)/(a − cA) provided that a 6= cA.
(c) Multiply the equation by x1/2 to obtain 12 p = wx1/2 . Thus x1/2 = p/2w and so, after
squaring each side, x = p2 /4w2 .
√
(d) Multiply each side by 1 + x to obtain 1 + x + ax = 0, so x = −1/(1 + a).
(e) x2 = b2 /a2 , so x = ±b/a provided that a 6= 0.
(f) We see immediately that x = 0.
4. (a) αx − a = βx − b if and only if (α − β)x = a − b, so x = (a − b)/(α − β) provided that
α 6= β.
√
(b) Squaring each side of pq = 3q + 5 yields pq = (3q + 5)2 , so p = (3q + 5)2 /q provided
that q 6= 0.
(c) Y = 94 + 0.2[Y − (20 + 0.5Y )] = 94 + 0.2Y − 4 − 0.1Y , so 0.9Y = 90, implying that
Y = 100.
r
K = Q4 , so K 3 = 2wQ4 /r, and
(d) Raise each side to the fourth power to obtain K 2
2w
1/3
hence K = 2wQ4 /r
.
(e) Multiplying the numerator and denominator of the left-hand fraction by 4K 1/2 L3/4 leads
to 2L/K = r/w, from which we get L = rK/2w.
(f) Raise each side to the fourth power to obtain
1 4 −3 −1
p r w .
K = 32
1 4 −1
(r/2w)
16 p K
= r4 . It follows that
3.3 Quadratic equations
3. These results are straightforward and can be found in the book. To illustrate the process,
√
√
√
let us consider part (d): First, rewrite the equation as r2 + ( 3 − 2)r − 6 = 0, and then
9
apply the formula to obtain
√
r=
−( 3 −
√
q√
√
√
2) ± ( 3 − 2)2 − 4 · 1 · (− 6)
q 2·1
√
√
√
√
1
=
− 3+ 2± 3−2 6+2+4 6
2
q
√
√
√
1
− 3+ 2± 3+2 6+2
=
2
q√
√
√
√
1
2
− 3 + 2 ± ( 3 + 2)
=
2
√
√
√ 1 √
− 3 + 2 ± ( 3 + 2) ,
=
2
√
√
√
√
1
so the solutions are r = 2 2 2 = 2 and r = − 21 2 3 = − 3.
5. (a) See the solution in the book.
(b) If the smaller of the two natural numbers is n, then the larger is n+1, so the requirement
is that n2 + (n + 1)2 = 13. This reduces to 2n2 + 2n − 12 = 0, i.e. n2 + n − 6 = 0, with
solutions n = −3 and n = 2, so the two numbers are 2 and 3.7
(c) If the shorter side has length x, then the other side has length x + 14. According to
Pythagoras’s Theorem one has x2 + (x + 14)2 = 342 , or x2 + 14x − 480 = 0. The only
positive solution is x = 16, and then x + 14 = 30.
(d) If the usual driving speed is s km/h and the usual time spent is t hours, then st = 80.
Since 16 minutes is 16/60 = 4/15 hours, driving at the speed s + 10 km/h for s − 4/15
hours gives (s + 10)(t − 4/15) = 80. From the first equation, t = 80/s. Inserting this
into the second equation, we get (s + 10)(80/s − 4/15) = 80. Rearranging, we obtain
s2 + 10s − 3000 = 0, whose only positive solution is s = 50. So his usual driving speed
is 50 km/h.
3.4 Nonlinear equations
2. (a) The numerator 5 + x2 is never 0, so there are no solutions.
x2 + 1 + 2x
(x + 1)2
= 0, or 2
= 0, so x = −1.
(b) The equation is obviously equivalent to
2
x +1
x +1
(c) Because x = −1 is clearly not a solution, we can multiply the equation by (x + 1)2/3 to
obtain the equivalent equation (x + 1)1/3 − 13 x(x + 1)−2/3 = 0. Multiplying this equation
again by (x + 1)2/3 yields x + 1 − 31 x = 0, whose solution is x = −3/2.
(d) Multiplying by x − 1 yields x + 2x(x − 1) = 0, or x(2x − 1) = 0. Hence x = 0 or x = 1/2.
3. (a) z = 0 satisfies the equation. If z 6= 0, then z − a = za + zb, or (1 − a − b)z = a. If
a + b = 1 we have a contradiction. If a + b 6= 1, then z = a/(1 − a − b).
(b) The equation is equivalent to (1 + λ)µ(x − y) = 0, so λ = −1, µ = 0, or x = y.
(c) µ = ±1 makes the equation meaningless. Otherwise, multiplying the equation by 1 − µ2
yields λ(1 − µ) = −λ, or λ(2 − µ) = 0, so λ = 0 or µ = 2.
7
(d) The equation is equivalent to b(1 + λ)(a − 2) = 0, so b = 0, λ = −1, or a = 2.
If we had asked for integer solutions, we would have −3 and −2 in addition.
10
3.5 Using implication arrows
3. (a) If
√
x−4=
√
x + 5 − 9,
(i)
then squaring each side gives
√
x − 4 = ( x + 5 − 9)2 .
(ii)
√
Expanding the square on the right-hand side of (ii) gives x − 4 = x + 5 − 18 x + 5 + 81,
√
√
which reduces to 18 x + 5 = 90 or x + 5 = 5, implying that x + 5 = 25 and so x = 20.
This hows that if x is a solution of (i), then x = 20. No other value of x can satisfy (i).
But if we check this solution, we find that with x = 20 the left-hand side of (i) becomes
√
√
16 = 4, and the right-hand side becomes 25 − 9 = 5 − 9 = −4. This means that
equation (i) actually has no solutions at all.8
y
y =9−
√
x+5
5
y=
5
√
x−4
10
20
15
-5
y=
√
x
25
x + 5− 9
Figure SM3.5.3
(b) If x is a solution of
√
x−4=9−
√
x + 5,
(iii)
then just as in part (a) we find that x must be a solution of
√
x − 4 = (9 − x + 5)2 .
(iv)
√
√
Now, (9 − x + 5)2 = ( x + 5 − 9)2 , so equation (iv) is equivalent to equation (ii) in
part (a). This means that (iv) has exactly one solution, namely x = 20. Inserting this
value of x into equation (iii), we find that x = 20 is a solution of (iii).
A geometric explanation of the results can be given with reference to Fig. SM3.5.3. We see
that the two solid curves in the figure have no point in common, that is, the expressions
√
√
x − 4 and x + 5 − 9 are not equal for any value of x.9 This explains why the equation
√
in (a) has no solution. The dashed curve y = 9 − x + 5, on the other hand, intersects
√
y = x + 5 for x = 20 (and only there), and this corresponds to the solution in part (b).10
8
But note that 42 = (−4)2 , i.e. the square of the left-hand side equals the square of the right-hand side. That is
how the spurious
√ solution
√ x = 20 managed to sneak in.
9
In fact, x − 4 − ( x + 5 − 9) increases with x, so there is no point of intersection farther to the right, either.
10
In part (a) it was necessary to check the result, because the transition from (i) to (ii) is only an implication,
not an equivalence. Similarly, it was necessary to check the result in part (b), since the transition from (iii) to (iv)
also is only an implication — at least, it is not clear that it is an equivalence. (Afterwards, it turned out to be an
equivalence, but we could not know that until we had solved the equation.)
11
3.6 Two linear equations in two unknowns
4. (a) If the two numbers are x and y, then x + y = 52 and x − y = 26. Adding the two
equations gives 2x = 78, so x = 39, and then y = 52 − 39 = 13.
(b) Let the cost of one table be $x and the cost of one chair $y. Then 5x + 20y = 1800 and
2x + 3y = 420. Solving this system yields x = 120, y = 60.
(c) Let x and y be the number of units produced of B and P, respectively. This gives the
equations x = 23 y and 200x + 300y = 180 000. Inserting the expression for x from the
first equation into the second equation gives 300y + 300y = 180 000, which yields the
solution y = 300 and then x = 450. Thus, 450 units of quality A and 300 units of quality
B should be produced.
(d) Suppose that the person invested $x at 5% and $y at 7.2%. Then x + y = 10 000 and
0.05x + 0.072y = 676. The solution is x = 2000 and y = 8000.
Review exercises for Chapter 3
2. (a) Assuming x 6= ±4, multiplying by the lowest common denominator (x−4)(x+4) reduces
the equation to (x − 3)(x + 4) = (x + 3)(x − 4) or x = −x, so x = 0.
(b) The given equation makes sense only if x 6= ±3. If we multiply the equation by the
common denominator (x+3)(x−3) we get 3(x+3)2 −2(x2 −9) = 9x+27 or x2 +9x+18 = 0,
with the solutions x = −6 and x = −3. The only solution of the given equation is
therefore x = −6.
(c) Subtracting 2x/3 from each side simplifies the equation to 0 = −1 + 5/x, whose only
solution is x = 5.
(d) Assuming x 6= 0 and x 6= ±5, multiply by the common denominator x(x − 5)(x + 5) to
get x(x − 5)2 − x(x2 − 25) = x2 − 25 − (11x + 20)(x + 5). Expanding each side of this
equation gives x3 − 10x2 + 25x − x3 + 25x = x2 − 25 − 11x2 − 75x − 100, which simplifies
to 50x = −125 − 75x with solution x = −1.
4. (a) Multiply each side of the equation by 5K 1/2 to obtain 15L1/3 = K 1/2 . Squaring each
side gives K = 225L2/3 .
(b) Raise each side to the power 1/t to obtain 1 + r/100 = 21/t , and so r = 100(21/t − 1).
(c) abx0b−1 = p, so x0b−1 = p/ab. Now raise each side to the power 1/(b − 1).
(d) Raise each side to the power −ρ to get (1 − λ)a−ρ + λb−ρ = c−ρ , or b−ρ = λ−1 [c−ρ −
(1 − λ)a−ρ ]. Now raise each side to the power −1/ρ.
4 Functions of One Variable
4.2 Basic definitions
1. (a) f (0) = 02 + 1 = 1, f (−1) = (−1)2 + 1 = 2, f (1/2) = (1/2)2 + 1 = 1/4 + 1 = 5/4, and
√
√
f ( 2) = ( 2)2 + 1 = 2 + 1 = 3.
12
(b) (i) Since (−x)2 = x2 , f (x) = f (−x) for all x. (ii) f (x+1) = (x+1)2 +1 = x2 +2x+1+1 =
x2 + 2x + 2 and f (x) + f (1) = x2 + 1 + 2 = x2 + 3. Thus equality holds if and only if
x2 + 2x + 2 = x2 + 3, i.e. if and only if x = 1/2. (iii) f (2x) = (2x)2 + 1p= 4x2 + 1 and
√
2f (x) = 2x2 + 2. Now, 4x2 + 1 = 2x2 + 2 ⇐⇒ x2 = 1/2 ⇐⇒ x = ± 1/2 = ± 21 2.
13. (a) We require 5 − x ≥ 0, so x ≤ 5.
(b) The denominator x2 − x = x(x − 1) must be different from 0, so x 6= 0 and x 6= 1.
(c) To begin with, the denominator must be nonzero, so we require x 6= 2 and x 6= −3.
Moreover, since we can only take the square root of a nonnegative number, the fraction
(x − 1)/(x − 2)(x + 3) must be ≥ 0. A sign diagram reveals that Df = (−3, 1] ∪ (2, ∞).
Note in particular that the function is defined with value 0 at x = 1.
4.4 Linear functions
10. The points that satisfy the inequality 3x + 4y ≤ 12 are those that lie on or below the straight
line 3x + 4y = 12, as explained in Example 4.4.6 for a similar inequality. The points that
satisfy the inequality x − y ≤ 1, or equivalently, y ≥ x − 1, are those on or above the straight
line x − y = 1. Finally, the points that satisfy the inequality 3x + y ≥ 3, or equivalently,
y ≥ 3− 3x, are those on or above the straight line 3x + y = 3. The set of points that satisfy all
these three inequalities simultaneously is the shaded set shown in Fig. A4.4.10 in the book.
4.6 Quadratic functions
9. The key to the argument is part (b). We find that f (x) = Ax2 + Bx + C, with coefficients
A = a21 + a22 + · · · + a2n , B = 2(a1 b1 + a2 b2 + · · · + an bn ), and C = b21 + b22 + · · · + b2n . Now, in case
B 2 −4AC > 0, then according to formula (2.3.4), the equation f (x) = Ax2 +Bx+C = 0 would
have two distinct solutions. This would contradict f (x) ≥ 0 for all x. Hence B 2 − 4AC ≤ 0
and the conclusion follows.
4.7 Polynomials
3. (a) The answer is 2x2 + 2x + 4 +
3
, because
x−1
(2x3
2x3 − 2x2
+ 2x − 1) ÷ (x − 1)= 2x2 + 2x + 4
2x2 + 2x − 1
2x2 − 2x
4x − 1
4x − 4
3
13
remainder
(b) The answer is x2 + 1, because
(x4 + x3 + x2 + x) ÷ (x2 + x) = x2 + 1
x4 + x3
x2 + x
x2 + x
0
(c) The answer is x3 − 4x2 + 3x + 1 −
4x
, because
x2 + x + 1
(x5 − 3x4
+ 1) ÷ (x2 + x + 1) = x3 − 4x2 + 3x + 1
x5 + x4 + x3
− 4x4 − x3
− 4x4 − 4x3 − 4x2
+1
3x3 + 4x2
+1
3
2
3x + 3x + 3x
x2 − 3x + 1
x2 + x + 1
−4x
remainder
(d) The answer is
3x5 + 6x3 − 3x2 + 12x − 12 +
because
(3x8
28x2 − 36x + 13
,
x3 − 2x + 1
+
x2
+ 1 ) ÷ (x3 − 2x + 1) = 3x5 + 6x3 − 3x2 + 12x − 12
6x6 − 3x5
+
6x6
− 12x4 + 6x3
x2
+ 1
−3x5 + 12x4 + 6x3 + x2
−3x5
+ 6x3 − 3x2
+ 1
3x8 − 6x6 + 3x5
12x4 − 12x3 + 4x2
+ 1
4
2
12x
− 24x + 12x
−12x3 + 28x2 − 12x + 1
−12x3
+ 24x − 12
28x2 − 36x + 13
remainder
4. (a) Since the graph intersects the x-axis at the two points x = −1 and x = 3, we try the
quadratic function f (x) = a(x + 1)(x − 3), for some constant a > 0. But the graph
passes through the point (1, −2), so we need f (1) = −2. Since f (1) = −4a for our chosen
function, a = 21 . This leads to the formula y = 12 (x + 1)(x − 3).
14
(b) Because the equation f (x) = 0 must have roots x = −3, 1, 2, we try the cubic function
f (x) = b(x + 3)(x − 1)(x − 2). Then f (0) = 6b. According to the graph, f (0) = −12. So
b = −2, and hence y = −2(x + 3)(x − 1)(x − 2).
(c) Here we try a cubic polynomial of the form y = c(x + 3)(x − 2)2 , with x = 2 as a double
root. Then f (0) = 12c. From the graph we see that f (0) = 6, and so c = 12 . This leads
to the formula y = 12 (x + 3)(x − 2)2 .
8. Polynomial division gives
(x2 −
2
x +
γx
) ÷ (x + β) = x − (β + γ)
βx
− (β + γ)x
− (β + γ)x−β(β + γ)
β(β + γ) remainder
and so
αβ(β + γ)
β(β + γ)
.
E = α x − (β + γ) +
= αx − α(β + γ) +
x+β
x+β
4.8 Power functions
4. (a) C. The graph is a parabola and since the coefficient in front of x2 is positive, it has a
minimum point.
√
(b) D. The function is defined for x ≤ 2 and crosses the y-axis at y = 2 2 ≈ 2.8.
(c) E. The graph is a parabola and since the coefficient in front of x2 is negative, it has a
maximum point.
(d) B. When x increases, y decreases, and y becomes close to −2 when x is large.
(e) A. The function is defined for x ≥ 2 and increases as x increases.
(f) F. Let y = 2 − ( 21 )x . Then y increases as x increases. For large values of x, one has y
close to 2.
4.10 Logarithmic functions
3. (a) 3x 4x+2 = 8 when 3x 4x 42 = 8 or (12)x 42 = 8, and so 12x = 1/2. Taking the natural log
of each side gives x ln 12 = − ln 2, so x = − ln 2/ ln 12.
(b) Since ln x2 = 2 ln x, the equation reduces to 7 ln x = 6, so ln x = 6/7, and thus x = e6/7 .
(c) One possible way to solve the equation is to rewrite it as 4x (1 − 4−1 ) = 3x (3 − 1), or
4x · (3/4) = 3x · 2, so (4/3)x = 8/3, implying that x = ln(8/3)/ ln(4/3). Alternatively,
start by dividing both sides by 3x to obtain (4/3)x (1 − 1/4) = 3 − 1 = 2, so (4/3)x = 8/3
as before.
For parts (d)–(f) below, we use the definition aloga x = x for all a, x > 0.
(d) log2 x = 2 implies that 2log2 x = 22 or x = 4.
2
(e) logx e2 = 2 implies that xlogx e = x2 or e2 = x2 . Hence x = e.
15
(f) log3 x = −3 implies that 3log3 x = 3−3 or x = 1/27.
4. Directly, the equation Aert = Best implies that ert /est = B/A, so e(r−s)t = B/A. Taking the
1
ln B . Alternatively, apply ln to each
ln of each side gives (r − s)t = ln(B/A), so t =
r−s A
side of the original equation, leading to ln A + rt = ln B + st. This gives the same solution.
5. Let t denote the number of years after 1990. Assuming continuous exponential growth, when
the GDP of the two nations is the same, one must have 1.2 · 1012 · e0.09t = 5.6 · 1012 · e0.02t .
Applying the answer found in part (a), we obtain
t=
5.6 · 1012
1
14
1
ln
=
ln
≈ 22
0.09 − 0.02 1.2 · 1012
0.07
3
According to this, the two countries would have the same GDP approximately 22 years after
1990, so in 2012.
Review exercises for Chapter 4
14. (a) p(x) = x(x2 + x − 12) = x(x − 3)(x + 4), because x2 + x − 12 = 0 for x = 3 and x = −4.
(b) ±1, ±2, ±4, ±8 are the only possible integer zeros. By trial and error we find that
q(2) = q(−4) = 0, so 2(x − 2)(x + 4) = 2x2 + 4x − 16 is a factor for q(x). By polynomial
division we find that q(x) ÷ (2x2 + 4x − 16) = x − 1/2, so q(x) = 2(x − 2)(x + 4)(x − 1/2).
15
11
15
17. Check by direct calculation that p(2) = 14 23 − 22 − 11
4 2 + 2 = 2 − 4 − 2 + 2 = 0, so x − 2
must be a factor of p(x). By direct division, we find that p(x) ÷ (x − 2) = 41 (x2 − 2x − 15),
which factors as 14 (x + 3)(x − 5), so x = −3 and x = 5 are the two other zeros. (Alternatively,
p(x) has the same zeros as q(x) = 4p(x) = x3 − 4x2 − 11x + 30. Any integer zero of q(x) must
be a factor of 30, so the only possibilities are ±1, ±2, ±3, ±5, ±10, ±15, and ±30. But it is
tedious to check all 14 possible values of x in order to find the zeros in this way, and it does
not find any noninteger roots.)
a + b/x
, so that y tends
1 + c/x
to a as x becomes large positive or negative. The graph shows that a > 0. There is a break
point at x = −c, and −c > 0, so c < 0. Also f (0) = b/c > 0, so b < 0. The right-hand graph
of the quadratic function g is a parabola which is convex, so p > 0. Moreover r = g(0) < 0.
Finally, g(x) has its minimum at x = x∗ = −q/2p. Since x∗ > 0 and p > 0, we conclude that
q < 0.
19. For the left-hand graph, note that for x 6= 0, one has y = f (x) =
23. (a) ln(x/e2 ) = ln x − ln e2 = ln x − 2 for x > 0.
(b) ln(xz/y) = ln(xz) − ln y = ln x + ln z − ln y for x, y, z > 0
(c) ln(e3 x2 ) = ln e3 + ln x2 = 3 + 2 ln x for x > 0. (In general, ln x2 = 2 ln |x|.)
(d) When x > 0, note that ln(1/x) = − ln x and so
1
2
ln x − 32 ln(1/x) − ln(x + 1) = 2 ln x − ln(x + 1) = ln x2 − ln(x + 1) = ln[x2 /(x + 1)]
16
5 Properties of Functions
5.3 Inverse functions
4. (a) The function f does have an inverse since it is one-to-one. This is shown in the table by
the fact that all the numbers in the second row, the domain of f −1 , are different. The
inverse assigns to each number in the second row, the corresponding number in the first
row. In particular, f −1 (2) = −1.
(b) Since f (x) increases by 2 for each unit increase in x, one has f (x) = 2x + a for a suitable
constant a. But then a = f (0) = 4, so f (x) = 2x + 4. Solving y = 2x + 4 for x yields
x = 12 y − 2, so interchanging x and y gives y = f −1 (x) = 12 x − 2.
9. (a) (x3 − 1)1/3 = y ⇐⇒ x3 − 1 = y 3 ⇐⇒ x = (y 3 + 1)1/3 . If we use x as the independent
variable, f −1 (x) = (x3 + 1)1/3 . R is the domain and range for both f and f −1 .
(b) The domain consists of all x 6= 2. For all such x one has
x+1
−2y − 1
2y + 1
= y ⇐⇒ x + 1 = y(x − 2) ⇐⇒ (1 − y)x = −2y − 1 ⇐⇒ x =
=
.
x−2
1−y
y−1
Using x as the independent variable, f −1 (x) = (2x + 1)/(x − 1). The domain of the
inverse is all x 6= 1.
(c) Here one has y = (1 − x3 )1/5 + 2 ⇐⇒ y − 2 = (1 − x3 )1/5 ⇐⇒ (y − 2)5 = 1 − x3 ⇐⇒
x3 = 1 − (y − 2)5 , which is equivalent to x = [1 − (y − 2)5 ]1/3 . With x as the free variable,
f −1 (x) = [1 − (x − 2)5 ]1/3 . For both f and f −1 , the whole of the real line R is both the
domain and the range.
10. (a) The domain is R and the range is (0, ∞), so the inverse is defined on (0, ∞). From
y = ex+4 one has ln y = x + 4, so x = ln y − 4, y > 0.
(b) The range is R, which is the domain of the inverse. From y = ln x−4, one has ln x = y +4,
and so x = ey+4 .
(c) The domain is R. On this domain the function is increasing, with y → ln 2 as x → −∞
and y → ∞ as x → ∞. So the range of the function is (ln 2, ∞). From y = ln 2 + ex−3
one has ey = 2 + ex−3 , so ex−3 = ey − 2. Hence, x = 3 + ln(ey − 2) for y > ln 2.
5.4 Graphs of equations
√
√
1. (a) The curve intersects the axes x = 0 and y = 0 at the points (0, ± 3) and (± 6, 0) respec√
√
tively. It is also entirely bounded by the rectangle whose four corners are (± 6, ± 3).
√
√
Moreover, it is symmetric about both axes, since all its points take the form (± ξ, ± η),
where ξ, η are any pair of positive real numbers satisfying ξ 2 + 2η 2 = 6. Putting x = y
√
√
yields the four points (± 2, ± 2) on the curve. More points can be found by fixing any
x satisfying −6 < x < 6, then solving for y.11
(b) The same argument as in (a) shows that the curve intersects only the axis x = 0, at
(0, ±1). There are no points on the graph where y 2 < 1. As in (a), it is symmetric about
11
The curve is called an ellipse. See the next section.
17
both axes. It comes in two separate parts: below y = −1; above y = 1. Putting x2 = 1
√
√
and then x2 = 9 yields the additional points (±1, ± 2) and (±3, ± 10).12
5.5 Distance in the plane
8. The method of completing the square used in problem 5 shows that
x2 + y 2 + Ax + By + C = 0 ⇐⇒ x2 + Ax + y 2 + By + C = 0
2
⇐⇒ x2 + Ax + 21 A + y 2 + By +
⇐⇒
x + 21 A
2
+ y + 12 B
2
2
1
B
2
1
= (A2 + B 2 − 4C)
4
1
= (A2 + B 2 − 4C).
4
Provided that A2 + B 2 > 4C, the last equation is that of a circle centred at (− 21 A, − 21 B) with
√
radius 12 A2 + B 2 − 4C. If A2 + B 2 = 4C, the graph consists only of the point (− 12 A, − 21 B).
For A2 + B 2 < 4C, the solution set is empty.
5.6 General functions
1. In each case, except (c), the rule defines a function because it associates with each member
of the original set a unique member in the target set. For instance, in (d), if the volume
V of a sphere is given, the formula V = 43 πr3 in the appendix implies that the radius is
r = (3V /4π)1/3 . But then the formula S = 4πr2 in the appendix gives the surface area.
Substituting for r in this formula gives S = 4π(3V /4π)2/3 = (36π)1/3 V 3/2 that expresses the
surface area of a sphere as a function of its volume.
Review exercises for Chapter 5
7. (a) The function f is defined and strictly increasing for ex > 2, i.e. x > ln 2. Its range is R
because f (x) → −∞ as x → ln 2+ , and f (x) → ∞ as x → ∞. From y = 3 + ln(ex − 2),
we get ln(ex − 2) = y − 3, and so ex − 2 = ey−3 , or ex = 2 + ey−3 , so x = ln(2 + ey−3 ).
Hence f −1 (x) = ln(2 + ex−3 ), x ∈ R.
(b) Note that f is strictly increasing. Moreover, e−λx → ∞ as x → −∞, and e−λx → 0 as
x → ∞. Therefore, f (x) → 0 as x → −∞, and f (x) → 1 as x → ∞. So the range of f ,
a
and therefore the domain of f −1 , is (0, 1). Because y = −λx
we get e−λx + a = a/y,
e
+a
so e−λx = a(1/y−1), or −λx = ln a+ln(1/y−1). Thus x = −(1/λ) ln a−(1/λ) ln(1/y−1),
and therefore the inverse is f −1 (x) = −(1/λ) ln a − (1/λ) ln(1/x − 1), with x ∈ (0, 1).
6 Differentiation
6.2 Tangents and derivatives
6. For parts (a)–(c) we set out the explicit steps of the recipe in given in the book; for parts
(d)–(f) we still follow the recipe, but express the steps more concisely.
12
The graph is a hyperbola. See the next section.
18
(a) (A) f (a + h) = f (0 + h) = 3h + 2; (B) f (a + h) − f (a) = f (h) − f (0) = 3h + 2 − 2 = 3h;
(C)–(D) [f (h) − f (0)]/h = 3; (E) [f (h) − f (0)]/h = 3 → 3 as h → 0, so f ′ (0) = 3. The
slope of the tangent at (0, 2) is 3.
(b) (A) f (a + h) = f (1 + h) = (1 + h)2 − 1 = 1 + 2h + h2 − 1 = 2h + h2 ;
(B) f (1 + h) − f (1) = 2h + h2 ; (C)–(D) [f (1 + h) − f (1)]/h = 2 + h;
(E) [f (1 + h) − f (1)]/h = 2 + h → 2 as h → 0, so f ′ (1) = 2.
(c) (A) f (3 + h) = 2 + 3/(3 + h); (B) f (3 + h) − f (3) = 2 + 3/(3 + h) − 3 = −h/(3 + h);
(C)–(D) [f (3 + h) − f (3)]/h = −1/(3 + h);
(E) [f (3 + h) − f (3)]/h = −1/(3 + h) → −1/3 as h → 0, so f ′ (3) = −1/3.
(d) [f (h) − f (0)]/h = (h3 − 2h)/h = h2 − 2 → −2 as h → 0, so f ′ (0) = −2.
1
1
1
h2 − 1 + 1
(e) [f (−1 + h) − f (−1)] =
−1 + h +
+ 2 , which simplifies to
=
h
h
−1 + h
h(h − 1)
h
. This tends to 0 as h → 0, so f ′ (0) = 0.
h−1
1
1
1
(f) [f (1+h)−f (1)] = [(1+h)4 −1] = [(h4 +4h3 +6h2 +4h+1)−1] = h3 +4h2 +6h+4 → 4
h
h
h
as h → 0, so f ′ (1) = 4.
√
√
9. (a) Applying the formula (a − b)(a + b) = a2 − b2 with a = x + h and b = x gives
√
√ √
√
( x + h − x)( x + h + x) = (x + h) − x = h.
(b) By direct computation,
√
√
√
√
1
f (x + h) − f (x)
( x + h − x )( x + h + x )
√
=√
=
√
√ .
h
h( x + h + x )
x+h+ x
(c) From part (b), as h → 0 one has
1
1
f (x + h) − f (x)
→ √ = x−1/2 .
h
2
2 x
6.5 A dash of limits
5. (a)
1/3 − 2/3h
3h (1/3 − 2/3h)
h−2
1
1
=
=
=
→ as h → 2.
h−2
3h(h − 2)
3h(h − 2)
3h
6
(b) When x → 0, then (x2 − 1)/x2 = 1 − 1/x2 → −∞ as x → 0.
r
√
32t − 96
32(t − 3)
32
32t − 96
=
=
→ 8, as t → 3, so 3 2
→ 3 8 = 2 as t → 3.
(c) 2
t − 2t − 3
(t − 3)(t + 1)
t+1
t − 2t − 3
√
√ √
√
√
√
( h + 3 − 3 )( h + 3 + 3 )
h+3− 3
1
1
√
√ → √ as h → 0.
√
=
(d)
=√
h
h( h + 3 + 3 )
h+3+ 3
2 3
(t + 2)(t − 2)
t−2
2
t2 − 4
=
=
→ − as t → −2.
+ 10t + 16
(t + 2)(t + 8)
t+8
3
√
√
√
1
2− x
1
√ = .
(f) Observe that 4 − x = (2 + x)(2 − x), so lim
= lim
x→4 4 − x
x→4 2 + x
4
(e)
t2
x2 + 2x − 3
(x − 1)(x + 3)
f (x) − f (1)
=
=
= x + 3 → 4 as x → 1
x−1
x−1
x−1
f (x) − f (1)
= x + 3 → 5 as x → 2.
(b)
x−1
6. (a)
19
(2 + h)2 + 2(2 + h) − 8
h2 + 6h
f (2 + h) − f (2)
=
=
= h + 6 → 6 as h → 0.
h
h
h
f (x) − f (a)
x2 + 2x − a2 − 2a
x2 − a2 + 2(x − a)
(x − a)(x + a) + 2(x − a)
(d)
=
=
=
,
x−a
x−a
x−a
x−a
which equals x + a + 2 and tends to 2a + 2 as x → a.
(c)
(e) Same answer as part (d), putting x − a = h.
(f)
f (a + h) − f (a − h)
(a + h)2 + 2a + 2h − (a − h)2 − 2a + 2h
=
= 4a + 4, which tends to
h
h
4a + 4 as a limit as h → 0.
6.7 Sums, products and quotients
3. (a) y =
(b) y =
1
= x−6 ⇒ y ′ = −6x−7 , using the power rule (6.6.4).
x6
√
x−1 (x2 + 1) x = x−1 x2 x1/2 + x−1 x1/2 = x3/2 + x−1/2
(c) y = x−3/2 ⇒ y ′ = − 23 x−5/2 .
⇒ y ′ = 32 x1/2 − 12 x−3/2 .
1 · (x − 1) − (x + 1) · 1
−2
x+1
⇒ y′ =
=
.
2
x−1
(x − 1)
(x − 1)2
x
1
4
5
(e) y = 5 + 5 = x−4 + x−5 ⇒ y ′ = − 5 − 6 .
x
x
x
x
3x − 5
3(2x
+
8)
−
2(3x
−
5)
34
(f) y =
⇒ y′ =
=
2
2x + 8
(2x + 8)
(2x + 8)2
(d) y =
(g) y = 3x−11 ⇒ y ′ = −33x−12 .
(h) y =
3(x2 + x + 1) − (3x − 1)(2x + 1)
−3x2 + 2x + 4
3x − 1
′
⇒
y
=
=
x2 + x + 1
(x2 + x + 1)2
(x2 + x + 1)2
6. (a) y ′ = 6x − 12 = 6(x − 2) ≥ 0 ⇐⇒ x ≥ 2, so y is increasing in [2, ∞).
√
√
(b) y ′ = x3 − 3x = x(x2 − 3) = x(x − 3 )(x + 3 ), so, using a sign diagram, one sees that
√
√
y is increasing in [− 3, 0] and in
3, ∞ .
√ √
(c) y is increasing in [− 2, 2], since
√
√
2(2 − x2 )
2( 2 − x)( 2 + x)
2(2 + x2 ) − (2x)(2x)
′
= 2
=
.
y =
(2 + x2 )2
(x + 2)2
(x2 + 2)2
(d) y is increasing in (−∞, x1 ] and in [0, x2 ], since
−2x3 − 2x2 + 2x
−x(x − x1 )(x − x2 )
(2x − 3x2 )(x + 1) − (x2 − x3 )
=
=
,
2
2
2(x + 1)
2(x + 1)
(x + 1)2
√
where x1,2 = − 12 ± 12 5.
y′ =
7. (a) y ′ = −1 − 2x = −3 when x = 1, so the slope of the tangent is −3. Since y = 1 when
x = 1, the point–slope formula gives y − 1 = −3(x − 1), or y = −3x + 4.
(b) y = 1 − 2(x2 + 1)−1 , so y ′ = 4x/(x2 + 1)2 = 1 and y = 0 when x = 1. The tangent is
therefore y = x − 1.
(c) y = x2 − x−2 , so y ′ = 2x + 2x−3 = 17/4. Also y = 15/4 when x = 2, so the point–slope
formula gives y − 15/4 = (17/4)(x − 2), or y = (17/4)x − 19/4.
20
1
4x3 (x3 + 3x2 + x + 3) − (x4 + 1)(3x2 + 6x + 1)
= − . Also y =
2
2
[(x + 1)(x + 3)]
9
x = 0, so the tangent is y = 13 − 19 x = −(x − 3)/9.
(d) Here y ′ =
1
3
when
at + b
a(ct + d) − (at + b)c
ad − bc
⇒ y′ =
=
.
9. (a) By the quotient rule, y =
2
ct + d
(ct + d)
(ct + d)2
√
(b) y = tn a t + b = atn+1/2 + btn ⇒ y ′ = (n + 1/2)atn−1/2 + nbtn−1 .
(c) y =
0 · (at2 + bt + c) − 1 · (2at + b)
−2at − b
1
′
⇒
y
=
=
.
2
2
2
at + bt + c
(at + bt + c)
(at2 + bt + c)2
6.8 The chain rule
1
= (x2 + x + 1)−5 = u−5 , where u = x2 + x + 1. The chain rule
+ x + 1)5
gives y ′ = (−5)u−6 u′ = −5(2x + 1)(x2 + x + 1)−6 .
p
√
√
(b) With u = x + x + x, y = u = u1/2 , so y ′ = 12 u−1/2 u′ . Now, u = x + v 1/2 , with
v = x + x1/2 . Then u′ = 1 + 12 v −1/2 v ′ , where v ′ = 1 + 12 x−1/2 . Thus, in the end,
3. (a) Write y =
(x2
y ′ = 21 u−1/2 u′ =
1
2
h
x + (x + x1/2 )1/2
i−1/2 h
i
1 + ( 21 (x + x1/2 )−1/2 (1 + 12 x−1/2 ) .
(c) See the answer in the book.
6.10 Exponential functions
4. (a) y ′ = 3x2 + 2e2x is obviously positive everywhere, so y is increasing in (−∞, ∞).
(b) y ′ = 10xe−4x +5x2 (−4)e−4x = 10x(1−2x)e−4x . A sign diagram shows that y is increasing
in [0, 1/2].
2
2
2
(c) y ′ = 2xe−x + x2 (−2x)e−x = 2x(1 − x)(1 + x)e−x . A sign diagram shows that y is
increasing in (−∞, −1] and in [0, 1].
6.11 Logarithmic functions
3. For most of these problems we need the chain rule. That is important in itself! But it implies
in particular that if u = f (x) is a differentiable function of x that satisfies f (x) > 0, then
d
u′
ln u = .
dx
u
1 1
1
1
=
.
(a) y = ln(ln x) = ln u with u = ln x implies that y ′ = u′ =
u
ln x x
x ln x
√
√
(b) y = ln 1 − x2 = ln u with u = 1 − x2 implies that
y′ =
√
1
−x
−2x
1 ′
√
u =√
=
2
2
u
1 − x2
1−x 2 1−x
1 − x2 = (1 − x2 )1/2 ⇒ y = 21 ln(1 − x2 ), and so on.
1
1
x
′
x
x
x
(c) y = e ln x ⇒ y = e ln x + e
= e ln x +
.
x
x
Alternatively,
21
(d) y =
3
ex
ln x2
⇒
y′
=
3
3x2 ex
(e) y = ln(ex + 1) ⇒ y ′ =
ln x2
ex
.
+1
(h) y = e2x
2 −x
2
1
3
2
2
x
3x ln x +
2x = e
.
x2
x
ex
2x + 3
.
+ 3x − 1
⇒ y ′ = −2ex (ex − 1)−2 .
(f) y = ln(x2 + 3x − 1) ⇒ y ′ =
(g) y = 2(ex − 1)−1
+
3
ex
x2
⇒ y ′ = (4x − 1)e2x
2 −x
.
5. (a) One must have x2 > 1, i.e. x > 1 or x < −1.
(b) ln(ln x) is defined when ln x is defined and positive, that is, for x > 1.
1
(c) The fraction
is defined when ln(ln x) is defined and different from 1. From
ln(ln x) − 1
(b), ln(ln x) is defined when x > 1. Further, ln(ln x) = 1 ⇐⇒ ln x = e ⇐⇒ x = ee . We
1
conclude that
is defined if and only if x > 1 with x 6= ee .
ln(ln x) − 1
6. (a) The function is defined for 4 − x2 > 0, that is in (−2, 2). f ′ (x) = −2x/(4 − x2 ) ≥ 0 in
(−2, 0], so this is where y is increasing.
(b) The function is defined for x > 0. f ′ (x) = x2 (3 ln x+1) ≥ 0 for ln x ≥ −1/3, or x ≥ e−1/3 ,
so y is increasing in [e−1/3 , ∞).
(c) The function is defined for x > 0, and
y′ =
2(1 − ln x)(−1/x)2x − 2(1 − ln x)2
(1 − ln x)(ln x − 3)
=
.
2
4x
2x2
A sign diagram reveals that y is increasing in x when 1 ≤ ln x ≤ 3 and so for x in [e, e3 ].
9. In these problems we can use logarithmic differentiation. Alternatively we can write the
functions in the form f (x) = eg(x) and then use the fact that f ′ (x) = eg(x) g ′ (x) = f (x)g ′ (x).
′
(x)
1
= 1 · ln(2x) + x · 2x
· 2 = ln(2x) + 1.
(a) Let f (x) = (2x)x . Then ln f (x) = x ln(2x), so ff (x)
′
x
Hence, f (x) = f (x)(ln(2x) + 1) = (2x) (ln x + ln 2 + 1).
√
√
√x
= e x ln x , so
(b) f (x) = x x = eln x
√ √
√
√
1
d √
ln x
x
′
x ln x
x
√ +
f (x) = e
·
( x ln x) = x
= x x− 2 ( 21 ln x + 1).
dx
x
2 x
√
(c) Here ln f (x) = x ln x = 21 x ln x, so f ′ (x)/f (x) = 12 (ln x + 1), which implies that f ′ (x) =
1 √ x
2 ( x) (ln x + 1).
11. (a) See the answer in the book.
(b) Let f (x) = ln(1 + x) − 21 x. Then f (0) = 0 and moreover f ′ (x) = 1/(x + 1) − 12 =
(1 − x)/2(x + 1). This is positive in (0, 1), so f (x) > 0 in (0, 1), thus establishing the
left-hand inequality. To prove the other inequality, put g(x) = x − ln(1 + x). Then
g(0) = 0 and g ′ (x) = 1 − 1/(x + 1) = x/(x + 1) > 0 in (0, 1), so the conclusion follows.
√
√
√
(c) Let f (x) = 2( x − 1) − ln x. Then f (1) = 0 and f ′ (x) = (1/ x) − 1/x = ( x − 1)/x,
which is positive for x > 1. The conclusion follows.
22
Review exercises for Chapter 6
15. (a) y ′ =
2
x
ln x ≥ 0 if x ≥ 1.
ex − e−x
≥ 0 ⇐⇒ ex ≥ e−x ⇐⇒ e2x ≥ 1 ⇐⇒ x ≥ 0
ex + e−x
(x − 1)(x − 2)
3x
=
≥ 0 ⇐⇒ x ≤ 1 or x ≥ 2, as a sign diagram will show.
(c) y ′ = 1 − 2
x +2
x2 + 2
(b) y ′ =
7 Derivatives in Use
7.1 Implicit differentiation
3. (a) Implicit differentiation w.r.t. x yields (∗) 1 − y ′ + 3y + 3xy ′ = 0. Solving for y ′ yields
y ′ = (1 + 3y)/(1 − 3x). The definition of the function implies that y = (x − 2)/(1 − 3x).
Substituting this in the expression for y ′ gives y ′ = −5/(1 − 3x)2 . Differentiating (∗)
w.r.t. x gives −y ′′ + 3y ′ + 3y ′ + 3xy ′′ = 0. Inserting y ′ = (1 + 3y)/(1 − 3x) and solving
for y ′′ gives y ′′ = 6y ′ /(1 − 3x) = −30/(1 − 3x)3 .
(b) Implicit differentiation w.r.t. x yields (∗) 5y 4 y ′ = 6x5 , so y ′ = 6x5 /5y 4 = (6/5)x1/5 .
Differentiating (∗) w.r.t. x gives 20y 3 (y ′ )2 + 5y 4 y ′′ = 30x4 . Inserting y ′ = 6x5 /5y 4 and
solving for y ′′ yields y ′′ = 6x4 y −4 − 4y −1 (y ′′ )2 = 6x4 y −4 − (144/25)x10 y −9 = (6/25)x−4/5 .
8. (a) y + xy ′ = g ′ (x) + 3y 2 y ′ , and solve for y ′ .
(b) g ′ (x + y)(1 + y ′ ) = 2x + 2yy ′ , and solve for y ′ .
(c) First, in order to differentiate g(x2 y) w.r.t. x, put z = g(u) where u = x2 y. It follows that
z ′ = g ′ (u)u′ where u′ = 2xy + x2 y ′ . So one has 2(xy + 1)(y + xy ′ ) = g ′ (x2 y)(2xy + x2 y ′ ).
Now solve for y ′ .
10. (a) Differentiate w.r.t. x while keeping in mind that y depends on x. This yields the equation
2(x2 + y 2 )(2x + 2yy ′ ) = a2 (2x − 2yy ′ ). Then solve for y ′ .
(b) Note that x = 0 would imply that y = 0. Excluding this possibility, we see that y ′ = 0
when x2 + y 2 = a2 /2, or y 2 = 21 a2 − x2 . Inserting this into the given equation yields
√
√
x = ± 41 a 6 and so y = ± 21 a 2. This yields the four points on the graph at which the
tangent is horizontal.
7.2 Economic examples
4. (a) Using (ii) and (iii) to substitute for C and Y respectively in equation (i), one has Y =
f (Y ) + I + X̄ − g(Y ).
(b) Differentiating w.r.t. I yields
dY /dI = f ′ (Y )(dY /dI) + 1 − g ′ (Y )(dY /dI) = (f ′ (Y ) − g ′ (Y ))(dY /dI) + 1
(∗)
Thus, dY /dI = 1/[1 − f ′ (Y ) + g ′ (Y )]. Imports should increase when income increases, so
g ′ (Y ) > 0. It follows that dY /dI > 0.
(c) Differentiating (∗) w.r.t. I yields, in simplified notation, d2 Y /dI 2 = (f ′′ − g ′′ )(dY /dI) +
(f ′ −g ′ )(d2 Y /dI 2 ), so d2 Y /dI 2 = (f ′′ −g ′′ )(dY /dI)/(1−f ′ +g ′ )2 = (f ′′ −g ′′ )/(1−f ′ +g ′ )3 .
23
7.3 Differentiating the inverse
5. (a) dy/dx = −e−x−5 , so dx/dy = 1/(dy/dx) = 1/(−e−x−5 ) = −ex+5 .
(b) dy/dx = −e−x /(e−x + 3), so dx/dy = −(e−x + 3)/e−x = −1 − 3ex
(c) Implicit differentiation w.r.t. x yields y 3 + x(3y 2 )(dy/dx) − 3x2 y − x3 (dy/dx) = 2. Solve
for dy/dx, and then invert.
7.4 Linear approximations
3. (a) f (0) = 1 and f ′ (x) = −(1 + x)−2 , so f ′ (0) = −1. Then f (x) ≈ f (0) + f ′ (0)x = 1 − x.
(b) f (0) = 1 and f ′ (x) = 5(1 + x)4 , so f ′ (0) = 5. Then f (x) ≈ f (0) + f ′ (0)x = 1 + 5x.
(c) f (0) = 1 and f ′ (x) = − 14 (1 − x)−3/4 , so f ′ (0) = − 14 . Then f (x) ≈ f (0) + f ′ (0)x = 1 − 41 x.
2
8. (a) We must differentiate 3xexy . To do so, first use the product rule in order to get
d
xy 2 = 3exy 2 + 3x d exy 2 . Then the chain rule gives d exy 2 = exy 2 (y 2 + x2yy ′ ). Overdx 3xe
dx
dx
2
2
all, therefore, implicit differentiation yields 3exy + 3xexy (y 2 + x2yy ′ ) − 2y ′ = 6x + 2yy ′ .
Putting x = 1 and y = 0 gives 3 − 2y ′ = 6, so y ′ = −3/2.
(b) y(x) ≈ y(1) + y ′ (1)(x − 1) = − 23 (x − 1)
7.5 Polynomial approximations
2. f ′ (x) = (1 + x)−1 , f ′′ (x) = −(1 + x)−2 , f ′′′ (x) = 2(1 + x)−3 , f (4) (x) = −6(1 + x)−4 ,
f (5) (x) = 24(1 + x)−5 . Then f (0) = 0, f ′ (0) = 1, f ′′ (0) = −1, f ′′′ (0) = 2, f (4) (0) = −6,
f (5) (0) = 24, and so
f (x) ≈ f (0) +
1 ′
1
1
1
1
f (0)x + f ′′ (0)x + f ′′′ (0)x3 + f (4) (0)x4 + f (5) (0)x5
1!
2!
3!
4!
5!
1
1
1
1
= x − x2 + x3 − x4 + x5 .
2
3
4
5
3. With f (x) = 5(ln(1 + x) −
√
1 + x) = 5 ln(1 + x) − 5(1 + x)1/2 we get
5
5
f ′ (x) = 5(1 + x)−1 − (1 + x)−1/2 and f ′′ (x) = −5(1 + x)−2 + (1 + x)−3/2
2
4
So f (0) = −5, f ′ (0) = 52 , and f ′′ (0) = − 15
4 . Hence, the Taylor polynomial of order 2 about
1 ′′
2
2
′
x = 0 is f (0) + f (0)x + 2 f (0)x = −5 + 25 x − 15
8 x .
7.6 Taylor’s formula
4. (a) We use Taylor’s formula (7.6.3) with g(x) = (1 + x)1/3 and n = 2. Then g ′ (x) =
(b)
1
−2/3 , g ′′ (x) = − 2 (1 + x)−5/3 , and g ′′′ (x) = 10 (1 + x)−8/3 , so g(0) = 1, g ′ (0) = 1 ,
3 (1 + x)
9
27
3
10
(1 + c)−8/3 . It follows that g(x) = 1 + 13 x − 19 x2 + R3 (x), where
g ′′ (0) = − 29 , g ′′′ (c) = 27
−8/3 x3 = 5 (1 + c)−8/3 x3 .
R3 (x) = 61 10
27 (1 + c)
81
c ∈ (0, x) and x ≥ 0, so (1 + c)−8/3 ≤ 1, and the inequality follows.
24
(c) Note that
√
3
1003 = 10(1 + 3 · 10−3 )1/3 . Using the approximation in part (a) gives
1
1
(1 + 3 · 10−3 )1/3 ≈ 1 + 3 · 10−3 − (3 · 10−3 )2 = 1 + 10−3 − 10−6 = 1.000999
3
9
√
So 3 1003 ≈ 10.00999. By part (b), the error R3 (x) in the approximation (1+3·10−3 )1/3 ≈
5
1.000999 satisfies |R3 (x)| ≤ 81
(3 · 10−3 )3 = 35 10−9 . Hence the error in the approximation
√
3
−9 < 2 · 10−8 , implying that the answer is correct
1003 ≈ 10.00999 is 10|R3 (x)| ≤ 50
3 10
to 7 decimal places.
7.7 Elasticities
x dA
= 0.
A dx
x ′
xf ′ xg ′
x
(b) Elx (f g) =
(f g)′ =
(f g + f g ′ ) =
+
= Elx f + Elx g.
fg
fg
f
g
′
f
xf ′ xg ′
x
f
xg gf ′ − f g ′
(c) Elx =
=
−
= Elx f − Elx g.
=
g
(f /g) g
f
g2
f
g
(d) See the answer in the book.
9. (a) Elx A =
(e) This case is like (d), but with +g replaced by −g, and +g ′ by −g ′ .
(f) z = f (u) and u = g(x) imply that
Elx z =
x u dz du
u dz x du
x dz
=
=
= Elu f (u) Elx u.
z dx
u z du dx
z du u dx
7.8 Continuity
3. By the properties of continuous functions stated in the book, all the functions are continuous
wherever they are defined. So (a) and (d) are defined everywhere. In (b) we must exclude
x = 1; in (c) the function is defined for x < 2. Next, in (e) we must exclude values of x that
make the denominator 0. These values satisfy x2 + 2x − 2 = 0, or (x + 1)2 = 3, so they are
√
x = ± 3 − 1. Finally, in (f), the first fraction requires x > 0, and then the other fraction is
also defined.
7.9 More on limits
2. (a) limx→0+ (x2 + 3x − 4) = 02 + 3 · 0 − 4 = −4.
x + |x|
x−x
(b) |x| = −x for x < 0. Hence, lim
= lim
= lim 0 = 0.
x
x
x→0−
x→0−
x→0−
x+x
x + |x|
= lim
= lim 2 = 2.
(c) |x| = x for x > 0. Hence, lim
+
+
x
x
x→0
x→0+
x→0
√
√
+
(d) As x → 0 one has x → 0 and so −1/ x → −∞.
(e) As x → 3+ one has x − 3 → 0+ and so x/(x − 3) → ∞.
(f) As x → 3− one has x − 3 → 0− , and so x/(x − 3) → −∞.
5. (a) Vertical asymptote, x = −1. Moreover, x2 ÷ (x + 1) = x − 1 + 1/(x + 1), so y = x − 1 is
an asymptote as x → ±∞.
25
(b) No vertical asymptote. Moreover. (2x3 −3x2 +3x−6)÷(x2 +1) = 2x−3+(x−3)/(x2 +1),
so y = 2x − 3 is an asymptote as x → ±∞.
(c) Vertical asymptote, x = 1. Moreover, (3x2 + 2x) ÷ (x − 1) = 3x + 5 + 5/(x − 1), so
y = 3x + 5 is an asymptote as x → ±∞.
(d) Vertical asymptote, x = 1. Moreover,
(5x4 − 3x2 + 1) ÷ (x3 − 1) = 5x + (−3x2 + 5x + 1)/(x3 − 1)
So y = 5x is an asymptote as x → ±∞.
√
9. This is rather tricky because the denominator is 0 at x1,2 = 2 ± 3. A sign diagram shows
that f (x) > 0 only in (−∞, 0) and in (x1 , x2 ). The text explains where f increases. See also
Fig. SM7.9.9.
y
6
f (x) =
4
3x
−x 2 + 4x − 1
2
−6 −4 −2
−2
2
4
6
8
x
−4
Figure SM7.9.9
7.10 The intermediate value theorem and Newton’s method
4. Recall from (4.7.6) that any integer root of the equation f (x) = x4 + 3x3 − 3x2 − 8x + 3 = 0
must be a factor of the constant term 3. The way to see this directly is to notice that we
must have
3 = −x4 − 3x3 + 3x2 + 8x = x(−x3 − 3x2 + 3x + 8)
and if x is an integer then the bracketed expression is also an integer. Thus, the only possible
integer solutions are ±1 and ±3. Trying each of these possibilities, we find that only −3 is
an integer solution.
There are three other real roots, with approximate values x0 = −1.9, y0 = 0.4, and z0 = 1.5. If
we use Newton’s method once for each of these roots we get the more accurate approximations
f (−1.9)
−0.1749
= −1.9 −
≈ −1.9 + 0.021 = −1.879
′
f (−1.9)
8.454
−0.4624
f (0.4)
= 0.4 −
≈ 0.4 − 0.053 = 0.347
y1 = 0.4 − ′
f (0.4)
−8.704
f (1.5)
−0.5625
z1 = 1.5 − ′
= 1.5 −
≈ 1.5 + 0.034 = 1.534
f (1.5)
16.75
x1 = −1.9 −
26
7.12 L’Hôpital’s rule
x−1
1
= “0/0” = lim
= 21 . Alternatively, use x2 − 1 = (x + 1)(x − 1).
x→1 2x
x2 − 1
x3 + 3x2 − 4
3x2 + 6x
6x + 6
lim 3
=
“0/0”
=
lim
= “0/0” = lim
= 3.
2
2
x→−2 x + 5x + 8x + 4
x→−2 3x + 10x + 8
x→−2 6x + 10
x4 − 4x3 + 6x2 − 8x + 8
4x3 − 12x2 + 12x − 8
lim
=
“0/0”
=
lim
= “0/0” =
x→2
x→2
x3 − 3x2 + 4
3x2 − 6x
2
12x − 24x + 12
, which equals 2.
lim
x→2
6x − 6
(1/x) − 1
(−1/x2 )
1
ln x − x + 1
lim
=
“0/0”
=
lim
=
“0/0”
=
lim
=− .
x→1 (x − 1)2
x→1 2(x − 1)
x→1
2
2
7
− 4
1
ln(7x + 1) − ln(4x + 4)
3
7x + 1
lim
ln
= lim
= “0/0” = lim 7x+1 4x+4 = .
x→1 x − 1
x→1
x→1
4x + 4
x−1
1
8
3. (a) lim
x→1
(b)
(c)
(d)
(e)
xx (ln x + 1) − 1
xx (ln x + 1)2 + xx (1/x)
xx − x
= “0/0” = lim
= “0/0” = lim
,
x→1
x→1
x→1 1 − x + ln x
−1 + 1/x
−1/x2
which equals −2, where we have used Example 6.11.4 to differentiate xx .
(f) lim
9. Here
1/g(x)
−1/(g(x))2 g ′ (x)
f (x)
= lim
= “0/0” = lim
·
x→a 1/f (x)
x→a −1/(f (x))2 f ′ (x)
x→a g(x)
(f (x))2 g ′ (x)
g ′ (x)
1
2
= lim
·
=
L
lim
= L2 lim ′
2
′
′
x→a (g(x))
x→a f (x)
x→a f (x)/g ′ (x)
f (x)
L = lim
The conclusion follows. (This argument ignores problems with “division by 0”, when either
f ′ (x) or g ′ (x) tends to 0 as x tends to a.)
Review exercises for Chapter 7
1+x
> 0, so the domain of f is the interval −1 < x < 1. As x → 1− , one
1−x
has f (x) → ∞; as x → −1− , one has f (x) → −∞. Since f ′ (x) = 1/(1 − x2 ) > 0 when
−1 < x < 1, f is strictly increasing and the range of f is R.
1 1+x
1+x
1+x
(b) From y = ln
one has ln
= 2y, so
= e2y . Then solve for x.
2 1−x
1−x
1−x
10. (a) We must have
12. (a) f ′ (x) = 2/(2x + 4) = (x + 2)−1 and f ′′ (x) = −(x + 2)−2 . We get f (0) = ln 4, f ′ (0) = 1/2,
and f ′′ (0) = −1/4, so f (x) ≈ f (0) + f ′ (0)x + 12 f ′′ (0)x2 = ln 4 + x/2 − x2 /8.
(b) g ′ (x) = −(1/2)(1 + x)−3/2 and g ′′ (x) = (3/4)(1 + x)−5/2 . We get g(0) = 1, g ′ (0) = −1/2,
and g ′′ (0) = 3/4, so g(x) ≈ 1 − x/2 + 3x2 /8.
(c) h′ (x) = e2x + 2xe2x and h′′ (x) = 4e2x + 4xe2x . We get h(0) = 0, h′ (0) = 1, and h′′ (0) = 4,
so h(x) ≈ x + 2x2 .
15. With x =
1
2
and n = 5, formula (7.6.6) yields
1
e2 = 1 +
1
2
1!
+
( 12 )2 ( 21 )3 ( 21 )4 ( 12 )5 ( 21 )6 c
+
+
+
+
e,
2!
3!
4!
5!
6!
27
where c is some number between 0 and 21 . Now,
( 1 )6
( 1 )6
1
1
≈ 0.00004340,
R6 ( ) = 2 e c < 2 2 =
2
6!
6!
23040
where we used the fact that c <
1
2
1
e2 ≈ 1 +
1!
+
1
2
1
implies ec < e 2 < 2. Thus, it follows that
( 12 )2 ( 21 )3 ( 12 )4 ( 21 )5
1 1
1
1
1
+
+
+
=1+ + +
+
+
≈ 1.6486979.
2!
3!
4!
5!
2 8 48 384 3840
1
Because the error is less than 0.000043, the approximation e 2 ≈ 1.649 is correct to 3 decimal
places.
23. (a) limx→3− (x2 − 3x + 2) = 9 − 9 + 2 = 2.
(b) Tends to +∞.
√
3 − x + 17
tends to +∞ as x → −1− , but to −∞ as x → −1+ , so there is no limit as
(c)
x+1
x → −1.
(d) Here
(2 − x)ex − x − 2
−ex + (2 − x)ex − 1
=
“0/0”
=
lim
= “0/0”
x→0
x→0
x3
3x2
−ex − ex + (2 − x)ex
−xex
−ex
1
= lim
= lim
= lim
=−
x→0
x→0 6x
x→0 6
6x
6
Note that by cancelling x at the penultimate step, we have avoided using l’Hôpital’s rule
a third time.
lim
(e) Here
lim
x→3
5
1
− 2
x−3 x −x−6
x2 − 6x + 9
= “0/0”
x→3 x3 − 4x2 − 3x + 18
1
2
2x − 6
= “0/0” = lim
=
= lim 2
x→3 6x − 8
x→3 3x − 8x − 3
5
= lim
x−4
1
1
= “0/0” = lim
=
. Alternatively, note that 2x2 − 32 = 2(x2 − 16) =
2
x→4 2x − 32
x→4 4x
16
2(x − 4)(x + 4), and then cancel x − 4.
(x − 2)(x − 1)
x2 − 3x + 2
=
= x − 1, which tends to 1 as x → 2.
(g) If x 6= 2, then
x−2
x−2
(h) If x 6= −1, then
√
√
√
4 − x + 17
16 − x − 17
−1
(4 − x + 17 )(4 + x + 17 )
√
√
√
=
=
=
2x + 2
(2x + 2)(4 + x + 17 )
(2x + 2)(4 + x + 17 )
2(4 + x + 17 )
(f) lim
1
which tends to − 16
as x → −1.
2
1
ln x 2
(ln x)
= lim
= 0, because of (7.12.3).
(i) lim
x→∞ 3x2
3 x→∞
x
√
√
24. When x → 0, the numerator tends to b − d and the denominator to 0, so the limit does
not exist when d 6= b. If d = b, however, then
1
√
√
−1/2 − 1 c(cx + b)−1/2
ax + b − cx + b
a−c
2 a(ax + b)
2
= “0/0” = lim
= √ .
lim
x→0
x→0
x
1
2 b
28
8 Single-variable Optimization
8.2 Simple tests for extreme points
2. Note that
√
√
8(3x2 + 4) − (8x)(6x)
8(2 − 3x)(2 + 3x)
h (x) =
=
,
(3x2 + 4)2
(3x2 + 4)2
√
√
so h has critical points at x1 = −2 3/3 and x2 = 2 3/3. A sign diagram shows that
h′ (x) < 0 in (−∞, x1 ) and in (x2 , ∞), whereas h′ (x) > 0 in (x1 , x2 ). Therefore h is strictly
decreasing in (−∞, x1 ], strictly increasing in [x1 , x2 ], and strictly decreasing again in [x2 , ∞).
√
Then, because h(x) → 0 as x → ±∞, it follows that the maximum of h occurs at x2 = 2 3/3
√
and the minimum at x1 = −2 3/3.
′
8. (a) y ′ = ex − 2e−2x and y ′′ = ex + 4e−2x . Hence y ′ = 0 when ex = 2e−2x , or e3x = 2, i.e.
x = 31 ln 2. Since y ′′ > 0 everywhere, the function is convex and this is a minimum point.
(b) y ′ = −2(x − a) − 4(x − b) = 0 when x =
y ′′ = −6.
1
3 (a
+ 2b). This is a maximum point since
(c) y ′ = 1/x − 5 = 0 when x = 15 . This is a maximum point since y ′′ = −1/x2 < 0 for all
x > 0.
10. (a) f ′ (x) = k − Aαe−αx = 0 when x = x0 = (1/α) ln(Aα/k). Note that x0 > 0 if and only
if Aα > k. Moreover f ′′ (x) = Aα2 e−αx > 0 for all x ≥ 0, so x0 solves the minimization
problem.
(b) Substituting for A in the answer to (a) gives the expression for the optimal height x0 .
Its value increases as p0 (probability of flooding) or V (cost of flooding) increases, but
decreases as δ (interest rate) or k (marginal construction cost) increases. The signs of
these responses are obviously what an economist would expect.
8.3 Economic examples
2. (a) π(Q) = Q(a−Q)−kQ = (a−k)Q−Q2 so π ′ (Q) = (a−k)−2Q = 0 for Q = Q∗ = 12 (a−k).
This maximizes π because π ′′ (Q) < 0. The monopoly profit is
1
1
1
π(Q∗ ) = −[ (a − k)]2 + (a − k) (a − k) = (a − k)2 .
2
2
4
(b) dπ(Q∗ )/dk = − 12 (a − k) = −Q∗ , as in Example 8.3.3.
(c) The new profit function is π̂(Q) = π(Q) + sQ = (a − k)Q − Q2 + sQ. Then π̂ ′ (Q) =
a − k − 2Q + s = 0 when Q̂ = 21 (a − k + s). Evidently Q̂ = 12 (a − k + s) = a − k provided
s = a − k, which is the subsidy required to induce the monopolist to produce a − k units.
8.4 The extreme value theorem
2. In all cases the maximum and minimum exist by the extreme value theorem. Follow the
recipe described in Example 8.4.1:
29
(a) f ′ (x) = −2 for all x in [0, 3], so the recipe tells us that both the maximum and minimum
points are at the ends of the interval [0, 3]. Since f (0) = −1 and f (3) = −7, the maximum
is at x = 0, the minimum at x = 3.13
(b) f (−1) = f (2) = 10 and f ′ (x) = 3x2 − 3 = 0 at x = ±1. The only critical point in the
interval [−1, 2] is x = 1, where f (1) = 6. There are two maxima at the endpoints, and a
minimum at x = 1.
(c) f (x) = x + 1/x, so f (1/2) = f (2) = 5/2 at the endpoints. Also, f ′ (x) = 1 − 1/x2 = 0 at
x = ±1. The only critical point in the interval [ 21 , 2] is x = 1, where f (1) = 2. There are
two maxima at the endpoints, and a minimum at x = 1.
√
(d) At the endpoints one has f (−1) = 4 and f ( 5) = 0. Because f ′ (x) = 5x2 (x2 − 3), there
√
√
are two critical points in the interval [−1, 5] at x = 0 and x = 3. The values at these
√
√
critical points are f (0) = 0 and f ( 3) = −6 3. The maximum is at x = −1 and the
√
minimum is at x = 3.
(e) f ′ (x) = 3x2 − 9000x + 6 · 106 = 3(x − 1000)(x − 2000) = 0 when x = 1000 and x = 2000.
At these critical points f (1000) = 2.5 · 109 and f (2000) = 2 · 109 . There is a minimum at
the endpoint x = 0 and a maximum at x = 3000.
6. (a) (f (2) − f (1))/(2 − 1) = (4 − 1)/1 = 3 and f ′ (x) = 2x, so f ′ (x) = 3 when x = x∗ = 3/2.
√
√
(b) (f (1) − f (0))/1 = (0 − 1)/1 = −1 and f ′ (x) = −x/ 1 − x2 = −1 when x = 1 − x2 .
Squaring each side of the last equation gives x2 = 1 − x2 and so x2 = 21 . This has two
√
√
solutions x = ± 12 2, of which only the positive solution satisfies x = 1 − x2 . So we
√
require x = x∗ = 21 2.
(c) (f (6) − f (2))/4 = −1/6 and f ′ (x) = −2/x2 = −1/6 when −12/x2 = −1 or x2 = 12, and
√
√
√
so x = ± 12. The required solution in [2, 6] is x = x∗ = 12 = 2 3.
√
√
√
(d) (f (4) − f (0))/4 = 1/4 = ( 25 − 9)/4 = (5 − 3)/4 = 1/2 and f ′ (x) = 21 2x/ 9 + x2 =
√
√
x/ 9 + x2 = 1/2 when 2x = 9 + x2 . Squaring each side of the last equation gives
√
4x2 = 9 + x2 and so 3x2 = 9. This has two solutions x = ± 3, of which only the positive
√
√
solution satisfies x/ 9 + x2 = 1/2. So we require x = x∗ = 3.
8.5 Further economic examples
4. (a) π(Q) = 1840Q − (2Q2 + 40Q + 5000) = 1800Q − 2Q2 − 5000. Since π ′ (Q) = 1800 − 4Q = 0
for Q = 450, and π ′′ (Q) = −4 < 0, it follows that Q = 450 maximizes profits.
(b) π(Q) = 2200Q − 2Q2 − 5000. Since π ′ (Q) = 2200 − 4Q = 0 for Q = 550, and π ′′ (Q) =
−4 < 0, it follows that Q = 550 maximizes profits.
(c) π(Q) = −2Q2 − 100Q − 5000. Here π ′ (Q) = −4Q − 100 < 0 for all Q ≥ 0, so the endpoint
Q = 0 maximizes profits.
8.6 Local extreme points
2. (a) This function is strictly decreasing, so it has no extreme points.
13
Actually the sign of f ′ (x) alone implies that the maximum is at the lower end of the interval, and the minimum
at the upper end.
30
(b) f ′ (x) = 3x2 − 3 = 0 for x = ±1. Because f ′′ (x) = 6x, we have f ′′ (−1) = −6 and
f ′′ (1) = 6, so x = −1 is a local maximum point, and x = 1 is a local minimum point.
(c) f ′ (x) = 1 − 1/x2 = 0 for x = ±1. With f ′′ (x) = 2/x3 , we have f ′′ (−1) = −2 and
f ′′ (1) = 2, so x = −1 is a local maximum point, and x = 1 is a local minimum point.
(d) f ′ (x) = 5x4 − 15x2 = 5x2 (x2 − 3) and f ′′ (x) = 20x3 − 30x. There are three critical points
√
√
√
√
at x = 0 and x = ± 3. Because f ′′ (0) = 0, whereas f ′′ (− 3) = −20 · 3 3 + 30 3 =
√
√
√
√
−30 3 < 0 and f ′′ ( 3) = 30 3 > 0, there is a local maximum at x = − 3 and a local
√
minimum at x = 3.
(e) This parabola has a local (and global) minimum at x = 3.
(f) f ′ (x) = 3x2 + 6x = 3x(x + 2) and f ′′ (x) = 6x + 6. There are two critical points at x = 0
and x = −2. Because f ′′ (0) = 6 and f ′′ (−2) = −6, there is a local maximum at x = −2
and a local minimum at x = 0.
y
8
6
4
2
−6 −4 −2
−2
2
4
6
8 10
x
−4
Figure SM8.6.3
3. See the graph in Fig. SM8.6.3.
(a) The function f (x) is defined if and only if x 6= 0 and x ≥ −6. We have f (x) = 0 at
x = −6 and at x = −2. At any other point x in the domain, f (x) has the same sign as
(x + 2)/x, so f (x) > 0 if x ∈ (−6, −2) or x ∈ (0, ∞).
(b) We first find the derivative of f :
f ′ (x) = −
x+2
2√
1
−4x − 24 + x2 + 2x
(x + 4)(x − 6)
√
√
√
x
+
6
+
=
=
2
x2
x 2 x+6
2x x + 6
2x2 x + 6
By means of a sign diagram we see that f ′ (x) > 0 if −6 < x < −4, f ′ (x) < 0 if
−4 < x < 0, f ′ (x) < 0 if 0 < x < 6, f ′ (x) > 0 if 6 < x. Hence, f is strictly decreasing
in [−4, 0) and in (0, 6], strictly increasing in [−6, −4] and in [6, ∞). It follows from the
first-derivative test (Thm. 8.6.1) that the two points x = −4 and x = 6 are respectively
√
√
√
a local maximum and a local minimum, with f (−4) = 12 2 and f (6) = 34 8 = 8 2/3.
Also, according to the definition (8.6.1), the point x = −6 is another local minimum
point.
√
(c) Since limx→0 x + 6 = 6 > 0, while limx→0− (1 + 2/x) = −∞ and limx→0+ (1 + 2/x) = ∞,
we see that limx→0− f (x) = −∞ and limx→0+ f (x) = ∞. Furthermore, as x → ∞,
2
1
1 1 12
1
1
x − 2x − 24
′
=
→ · 0 = 0.
·√
− − 2 ·√
f (x) =
2
2x
2 x x
2
x+6
x+6
31
7. f (x) = x3 + ax + b → ∞ as x → ∞, and f (x) → −∞ as x → −∞. By the intermediate value
theorem, the continuous function f has at least one real root. We have f ′ (x) = 3x2 + a. We
consider two cases.
First, in case a ≥ 0, one has f ′ (x) > 0 for all x 6= 0, so f is strictly increasing, and there is
only one real root. Note that 4a3 + 27b2 ≥ 0 in this case.
p
√
Second, in case a < 0, one has f ′ (x) = 0 for x = ± −a/3 = ± p, where p = −a/3. Because
√
√
f ′′ (x) = 3x, the function f has a local maximum at x = − p, where y = b + 2p p, and a
√
√
local minimum at x = p, where y = b − 2p p. If either of these local extreme values is
0, the equation has a double root, which is the case if and only if 4p3 = b2 , that is, if and
only if 4a3 + 27b2 = 0. Otherwise, the equation has three real roots if and only if the local
maximum value is positive and the local minimum value is negative. This occurs if and only
√
√
√
if both b > −2p p and b < 2p p. Equivalent conditions are |b| < 2p p ⇐⇒ b2 < 4p3 ⇐⇒
4a3 + 27b2 < 0.
8.7 Inflection points
3. The answers given in the text can be found in a straightforward way by considering the signs
of the following derivatives:
(a) y ′ = −e−x (1 + x), y ′′ = xe−x .
(b) y ′ = (x − 1)/x2 , y ′′ = (2 − x)/x3 .
(c) y ′ = x2 e−x (3 − x), y ′′ = xe−x (x2 − 6x + 6).
(d) y ′ =
(e) y ′ =
1−2 ln x
, y ′′ = 6 lnxx−5
.
4
x3
x
x
′′
x
2e (e − 1), y = 2e (2ex
− 1).
(f) y ′ = e−x (2 − x2 ), y ′′ = e−x (x2 − 2x − 2).
Review exercises for Chapter 8
8. (a) The answer given in the text is easily found from the derivative h′ (x) =
ex (2 − e2x )
.
(2 + e2x )2
(b) The function h is strictly increasing in (−∞, 12 ln 2], which includes (−∞, 0]. Also h(x) →
0 as x → −∞, and h(0) = 1/3. Thus, h defined on (−∞, 0] has an inverse defined on
ex
(0, 1/3] with values in (−∞, 0]. To find the inverse, note that
= y if and only if
2 + e2x p
y(ex )2 −ex +2y = 0. This quadratic equation in ex has the roots ex = (1 ± 1 − 8y 2 ])/2y.
We require the solution to satisfy x ≤ 0 and so ex ≤ 1 when 0 < y < 1/3. Now, taking
x
the positive
psquare root would give e > 1/2y > 6 when
p 0 < y < 1/3. So we must have
x
2
e = (1 − 1 − 8y )/2y, implying that x = ln(1 − 1 − 8y 2 ) − ln(2y). Using x as the
√
free variable gives the inverse function h−1 (x) = ln(1 − 1 − 8x2 ) − ln(2x). The function
and its inverse are graphed in Fig. SM8.R.8.
√
10. (a) Because 3 u is defined for all real u, the only points x not in the domain are the ones
√
√
for which x2 − a = 0, or x = ± a. So the domain of f consists of all x 6= ± a. Since
√
√
3
u > 0 if and only if u > 0, the denominator in the expression for f (x), 3 x2 − a, is
32
y
0.5
h
0.5
−1.5 −1.0 −0.5
x
−0.5
h−1
−1.0
−1.5
Figure SM8.R.8
√
√
positive if and only if x2 > a, i.e. if and only if x < − a or x > a. The numerator
in the expression for f (x) is x, and a sign diagram then reveals that f (x) is positive in
√
√
(− a, 0) and in ( a, ∞). he The graph of f is symmetric about the origin because
(b) Writing
√
3
x
−x
√
=
−
= −f (x),
f (−x) = p
3
3
x2 − a
(−x)2 − a
x2 − a as (x2 − a)1/3 and then differentiating yields
f ′ (x) =
1 2
x2 − a − x · 13 2x
1 · (x2 − a)1/3 − x · 13 (x2 − a)−2/3 · 2x
3 (x − 3a)
=
=
,
(x2 − a)2/3
(x2 − a)4/3
(x2 − a)4/3
Here the second equality was obtained by multiplying both denominator and numerator
√
by (x2 − a)2/3 . Of course, f ′ (x) is not defined at ± a. Except at these points, the
denominator is always positive (since (x2 − a)4/3 = ((x2 − a)1/3 )4 ). The numerator,
√
√
√
√
1 2
= 13 (x + 3a)(x − 3a), is 0 at x = ± 3a, nonnegative in (−∞, − 3a]
3 (x − 3a)
√
√
and in [ 3a, ∞). Since f and f ′ are not defined at ± a, we find that f (x) is increas√
√
√
√
√ √
ing in (−∞, − 3a] and in [ 3a, ∞), decreasing in [− 3a, − a), in (− a, a), and in
√
√
√
√
( a, 3a]. It follows that x = − 3a is a local maximum point and x = 3a is a local
minimum point.
(c) Differentiating once more, we find that
f ′′ (x) =
2
2
3 x(x
2
2
− a)4/3 − 31 (x2 − 3a) · 34 (x2 − a)1/3 · 2x
9 x(9a − x )
=
.
(x2 − a)8/3
(x2 − a)7/3
Here the second equality was obtained by dividing each term in the denominator and the
numerator by (x2 − a)1/3 , then simplifying the numerator. The resulting expression for
√
√
f ′′ (x) shows that there are inflection points where x equals −3 a, 0, and 3 a.14
11. Note first that f (x) → 0 as x → ±∞, as can be shown by dividing both numerator and
denominator by x3 , then taking limits. Next, differentiation yields
f ′ (x) =
14
18x2 (x4 + x2 + 2) − 6x3 (4x3 + 2x)
−6x2 (x4 − x2 − 6)
−6x2 (x2 − 3)(x2 + 2)
=
=
,
(x4 + x2 + 2)2
(x4 + x2 + 2)2
(x4 + x2 + 2)2
f ′′ (x) is 0 at these points, and changes sign around each of them.
33
√
so f has critical points at x = 0 and at x = ± 3. Moreover, f ′ changes sign from negative to
√
positive as x increases through − 3, then it switches back to negative as x increases through
√
√
√
3. It follows that x = 3 is a local (and global) maximum point, that x = − 3 is a local
(and global) minimum point, and x = 0 is neither. (It is an inflection point.) Note moreover
that f (−x) = −f (x) for all x, so the graph is symmetric about the origin. The graph of f is
shown in Fig. A8.R.11 in the book.
9 Integration
9.1 Indefinite integrals
R
R
R
(t3 + 2t − 3) dt = t3 dt + 2t dt − 3 dt = 14 t4 + t2 − 3t + C.
R
R
d
(x − 1)3 =
(b) (x − 1)2 dx = (x2 − 2x + 1) dx = 31 x3 − x2 + x + C. Alternatively, since dx
R
3(x − 1)2 , we have (x − 1)2 dx = 13 (x − 1)3 + C1 . This agrees with the first answer, with
C1 = C + 1/3.
R
R
(c) (x − 1)(x + 2) dx = (x2 + x − 2) dx = 31 x3 + 12 x2 − 2x + C.
3. (a)
R
(d) First evaluate (x + 2)3 = x3 + 6x2 + 12x + 8, then integrate term by term to get
R
R
(x + 2)3 dx = 41 x4 + 2x3 + 6x2 + 8x + C. Or more directly, check that (x + 2)3 =
1
4
4 (x + 2) + C1 .
R
(e) (e3x − e2x + ex ) dx = 31 e3x − 21 e2x + ex + C.
Z 3
Z x − 3x + 4
4
1
2
(f)
x −3+
dx =
dx = x3 − 3x + 4 ln |x| + C
x
x
3
4. (a) First simplify the integrand to obtain
(y − 2)2
y 2 − 4y + 4
=
= y 3/2 − 4y 1/2 + 4y −1/2 .
√
√
y
y
Next, integrate term by term to get
Z
Z
(y − 2)2
2
8
dy = (y 3/2 − 4y 1/2 + 4y −1/2 ) dy = y 5/2 − y 3/2 + 8y 1/2 + C.
√
y
5
3
(b) By polynomial division, the integrand is
Z
x3
1
= x2 − x + 1 −
. It follows that
x+1
x+1
x3 x2
x3
dx =
−
+ x − ln |x + 1| + C.
x+1
3
2
R
d
1
(1 + x2 )16 + C.
(1 + x2 )16 = 16(1 + x2 )15 · 2x = 32x(1 + x2 )15 , so x(1 + x2 )15 dx = 32
dx
R
11. f ′ (x) = (x−2 +x3 +2) dx = −x−1 + 14 x4 +2x+C. With f ′ (1) = 41 we have 41 = −1+ 14 +2+C,
so C = −1. Now integration yields
Z x4
x5
−1
−x +
f (x) =
+ 2x − 1 dx = − ln x +
+ x2 − x + D.
4
20
(c)
With f (1) = 0 we have 0 = − ln 1 +
1
20
1
+ 1 − 1 + D, so D = − 20
.
34
9.2 Area and definite integrals
5. We consider only parts (c) and (f).
Z 3
3
3
1 3
1
1
27 32
5
1 2 1 3
(c)
( x3 − x4 ) =
x − x dx =
x (2 − x) = − +
= .
3
12
12 12
12
−2 2
−2 6
−2 12
Z 3
3
1
9 4
5
1
ln(t − 1) + t2 = ln 2 + − = ln 2 + .
(f)
+ t dt =
t−1
2
2 2
2
2
2
√
√
6. (a) f (x) = x3 −3x2 +2x and so f ′ (x) = 3x2 −6x+2 = 0 for x0 = 1− 3/3 and x1 = 1+ 3/3.
We see that f ′ (x) > 0 ⇐⇒ x < x0 or x > x1 . Also, f ′ (x) < 0 ⇐⇒ x0 < x < x1 . So f
is (strictly) increasing in (−∞, x0 ] and in [x1 , ∞), but (strictly) decreasing in [x0 , x1 ].
(b) See the graph in the text. The integral is
Z
1
f (x) dx =
0
Z
1
0
(x3 − 3x2 + 2x) dx =
1
0
( 41 x4 − x3 + x2 ) =
1
4
−0=
1
4
9.3 Properties of indefinite integrals
4. Directly, one has
Z
1
(xp+q + xp+r ) dx =
0
1
0
xp+q+1
xp+r+1
1
1
+
=
+
p+q+1 p+r+1
p+q+1 p+r+1
5. Equality (a) implies a + b = 6. Also, f ′′ (x) = 2ax + b, so equality (b) implies 2a + b = 18.
R
It follows that a = 12 and b = 6, so f ′ (x) = 12x2 − 6x. But then f (x) = (12x2 − 6x) dx =
R2
4x3 − 3x2 + C, and since we want 0 (4x3 − 3x2 + C) = 18, we must have 16 − 8 + 2C = 18,
hence C = 5.
6. (a) See the answer given in the book.
Z 1
Z 1
1
1
4
83
2
2
(x4 + 4x2 + 4) dx = ( x5 + x3 + 4x) = .
(x + 2) dx =
(b)
3
15
0 5
0
0
√
2
1/2
x +x+ x+1
x(x + 1) + (x + 1)
(c) Note that
=
= x + (x + 1)−1/2 , and so
x+1
x+1
√
Z 1 2
1
√
√
x +x+ x+1
dx = ( 12 x2 + 2(x + 1)1/2 ) = 12 + 2 2 − 2 = 2 2 − 23 .
x+1
0
0
(d) Clearing fractions, A
x+b d
x+c+b−c d
A(b − c) d
+ =A
+ = A+
+ . Now integrate.
x+c x
x+c
x
x+c
x
11. W (T ) = K(1 − e−̺T )/̺T . Here W (T ) → 0 as T → ∞, and by l’Hôpital’s rule, W (T ) → K
as T → 0+ . For T > 0, we find W ′ (T ) = Ke−̺T (1 + ̺T − e̺T )/̺T 2 < 0 because e̺T > 1 + ̺T
(see Problem 6.11.11). We conclude that W (T ) is strictly decreasing and that W (T ) ∈ (0, K).
2
√
> 0 for all x > 0. Also, f (x) → −∞ as x → 0, whereas
x + 4 ( x + 4 − 2)
f (x) → ∞ as x → ∞. It follows that f is strictly increasing on (0, ∞), with range
equal to R. Hence f has an inverse defined on R. To find the inverse, note that y =
√
√
√
4 ln( x + 4 − 2) ⇐⇒ ln( x + 4 − 2) = y/4 ⇐⇒
x + 4 = ey/4 + 2 ⇐⇒ x + 4 =
y/4
2
y/2
y/4
(e + 2) ⇐⇒ x = e + 4e . It follows that the inverse is g(x) = ex/2 + 4ex/4 .
12. (a) f ′ (x) = √
35
(b) See Fig. A9.3.12.
(c) In Fig. A9.3.12 the graphs of f and g are symmetric about the line y = x, so the areas
of A and B are equal. But the area of B is the difference between: (i) the area of a
rectangle with base a and height 10; (ii) the area below the graph of g over the interval
[0, a]. Therefore,
Z a
A = B = 10a −
(ex/2 + 4ex/4 ) dx = 10a − 2ea/2 − 16ea/4 + 2 + 16
0
√
√
√
√
Because a = f (10) = 4 ln( 14−2), we have ea/2 = ( 14−2)2 = 14−4 14+4 = 18−4 14
√
and also ea/4 = 14 − 2. Hence,
√
√
√
√
A = B = 10a − 2(18 − 4 14) − 16( 14 − 2) + 18 = 40 ln( 14 − 2) + 14 − 8 14 ≈ 6.26
9.4 Economic applications
2. (a) Let n be the total number of individuals. The number of individuals with income in the
R 2b
2b
interval [b, 2b] is then N = n b Br−2 dr = n b −Br−1 = nB
2b .
R 2b
R 2b
2b
−2
−1
Their total income is M = n b Br r dr = n b Br dr = n b B ln r = nB ln 2.
Hence the mean income is m = M/N = 2b ln 2.
R 2b
R 2b
R 2b
(b) Total demand is x(p) = b nD(p, r)f (r) dr = b nApγ rδ Br−2 dr = nABpγ b rδ−2 dr
Evaluating the integral gives
Z 2b
2b δ−1
2δ−1 − 1
r
γ
= nABpγ bδ−1
x(p) = nABp
rδ−2 dr = nABpγ
δ−1
b
b δ−1
9.5 Integration by parts
1. (a) See the answer given in the book.
R
R
3 4x
(b) 3xe4x dx = 3x · 14 e4x − 3 · 41 e4x dx = 34 xe4x − 16
e + C.
R
R
R
2
−x
2
−x
−x
(c) (1 + x )e dx = (1 + x )(−e ) − 2x(−e ) dx = −(1 + x2 )e−x + 2 xe−x dx. Using
the answer to (a) to evaluate the last integral, we get
Z
(1 + x2 )e−x dx = −(1 + x2 )e−x − 2xe−x − 2e−x + C = −(x2 + 2x + 3)e−x + C
(d)
R
x ln x dx = 21 x2 ln x −
Z
2
R
1 21
2x x
dx = 21 x2 ln x −
R
1
2 x dx
= 21 x2 ln x − 14 x2 + C.
2. (a) See the answer given in the book.
d x
2 = 2x ln 2, so 2x / ln 2 is the indefinite integral of 2x . It follows that
(b) Recall that
dx
2
x2x dx =
0
x
0
2x
−
ln 2
Z
2
0
2x
8
dx =
−
ln 2
ln 2
2
0
4
8
1
3
8
2x
=
−
−
−
.
=
2
2
2
(ln 2)
ln 2
(ln 2)
(ln 2)
ln 2 (ln 2)2
(c) First use integration by parts on the indefinite integral. By Eq. (9.5.1) with f (x) = x2
and g(x) = ex ,
Z
Z
2 x
2 x
x e dx = x e − 2xex dx.
(∗)
36
To evaluate the last integral we must use integration by parts once more. With f (x) = 2x
R
R
and g(x) = ex , we get 2xex dx = 2xex − 2ex dx = 2xex − (2ex + C). Inserted into (∗),
R
this gives x2 ex dx = x2 ex − 2xex + 2ex + C. So
Z 1
1
x2 ex dx = (x2 ex − 2xex + 2ex ) = (e − 2e + 2e) − (0 − 0 + 2) = e − 2
0
0
Alternatively and more directly, use formula (9.5.2) to obtain
1
Z 1
Z 1
Z 1
1
2 x
x
x
2 x
x
xe dx = e−2
e dx = e−2 e −
xe −
x e dx = x e −2
0
0
0
0
0
1
e
x
0
= e−2.
(d) We must write the integrand in the form f (x)g ′ (x). If we let f (x) = x and g ′ (x) =
√
1 + x = (1 + x)1/2 , then what is g? A certain amount of reflection should suggest that
it is worth trying g(x) = 23 (1 + x)3/2 . Using (9.5.2) then gives
Z 3
Z 3
3
√
2
2
2
2 32
3/2
x 1 + x dx = x · (1 + x) −
1 · (1 + x)3/2 dx = 3 · · 43/2 −
(1 + x)5/2
3
3
3
3 05
0
0
0
R3 √
4
11
4
(45/2 − 1) = 16 − 15
· 31 = 7 15
. Alternatively, we
It follows that 0 x 1 + x dx = 16 − 15
√
could have found the indefinite integral of x 1 + x first, and then evaluated the definite
integral by using definition (9.2.3) of the definite integral.
√
Figure SM9.5.2(d) shows the area under the graph of y = x 1 + x over the interval [0, 3].
11
You should ask yourself whether 7 15
is a reasonable estimate of the area of A.
y
√
y =x 1+x
6
5
4
3
2
A
1
1
2
3
4
5
x
Figure SM9.5.2(d)
6. (a) By formula (9.5.2),
Z
T
T
te
0
−rt
dt =
0
−1
t e−rt −
r
Z
T
0
−1 −rt
−T −rT 1
e dt =
e
+
r
r
r
Z
T
e−rt dt
0
−T −rT 1 T −1 −rt
1
=
e
+
e
= 2 (1 − (1 + rT )e−rT ). Multiply this expression by b.
r
r 0 r
r
RT
R T −rt
RT
−rt
(b) 0 (a + bt)e dt = a 0 e dt + b 0 te−rt dt and so on using the result of part (a).
RT
RT
RT
RT
(c) 0 (a − bt + ct2 )e−rt dt = a 0 e−rt dt − b 0 te−rt dt + c 0 t2 e−rt dt. Now use the results
of parts (a) and (b) together with
Z T
Z
Z T T
1 −rt
1 2 −rT 2 T −rt
1 −rt
2 −rt
2
t e dt = t −
+
te dt.
e
−
e dt = − T e
2t −
r
r
r
r 0
0
0
0
37
9.6 Integration by substitution
2. (a) See the answer given in the book.
(b) With u = x3 +2 we get du = 3x2 dx and
R
x2 ex
3 +2
dx =
R
1 u
3 e du
= 13 eu +C = 13 ex
3 +2
+C.
(c) A first try might be u = x + 2, which gives du = dx and
Z
Z
ln(x + 2)
ln u
dx =
du.
2x + 4
2u
This, however, looks no simpler than the original integral.
A better idea is to substitute u = ln(x + 2). Then du =
Z
ln(x + 2)
dx =
2x + 4
Z
1
dx and
x+2
2
1
1
1
u du = u2 + C = ln(x + 2) + C.
2
4
4
(d) A first possible substitution is u = 1 + x, which gives du = dx and then
Z
Z
Z
√
√
x 1 + x dx = (u − 1) u du = (u3/2 − u1/2 ) du
This can be integrated to give
Z
√
2
2
2
2
x 1 + x dx = u5/2 − u3/2 + C = (1 + x)5/2 − (1 + x)3/2 + C
5
3
5
3
√
A second possible substitution is u = 1 + x. Now u2 = 1 + x and 2u du = dx. The
integral becomes
Z
Z
Z
√
x 1 + x dx = (u2 − 1)u · 2u du = (2u4 − 2u3 ) du
and so on. Check that you get the same answer.15
(e) With u = 1 + x2 one has x2 = u − 1, and du = 2x dx, so
Z
x3
dx =
(1 + x2 )3
x2 · x
dx =
(1 + x2 )3
Z
1
2
Z
u−1
du =
u3
1
2
Z
(u−2 − u−3 ) du
= − 21 u−1 + 14 u−2 + C = −
(f) With u =
√
1
1
+
+C
2(1 + x2 ) (1 + x2 )2
4 − x3 one has u2 = 4 − x3 and 2u du = −3x2 dx, so
2
u du
= x 4−
= (4 − u )u −
x 4−
3
Z 8
2
8
2
8
2
− u2 + u4 du = − u3 + u5 + C = − (4 − x3 )3/2 + (4 − x3 )5/2 + C
=
3
3
9
15
9
15
Z
5
p
x3 dx
Z
3
p
x3 x2 dx
Z
2
R1
R1
6. (a) I = 0 (x4 − x9 )(x5 − 1)12 dx = 0 −x4 (x5 − 1)13 dx. Introduce u = x5 − 1. Then
du = 5x4 dx. Now we use (9.6.2) along with the facts that u = −1 when x = 0 and u = 0
R0
0 1 14
1
when x = 1. The integral becomes I = − −1 51 u13 du = − −1 70
u = 70
.
15
Actually, even integration by parts works in this case. See Problem 9.5.2(d).
38
(b) With u =
√
x one has u2 = x and 2u du = dx. Then,16
ln x
√ dx = 2
x
Z
Z
2
ln u du = 4
Z
ln u du = 4(u ln u − u) + C
√
√
√
√
√
= 4 x ln x − 4 x + C = 2 x ln x − 4 x + C
√
(c) With u = 1 + x one has (u − 1)2 = x, so 2(u − 1) du = dx. Again we use (9.6.2) along
with the facts that u = 1 when x = 0 and u = 3 when x = 4. The specified integral
becomes17
Z 3
Z 3
8
3
2(u − 1)
1/2
−1/2
1/2
2 3/2
√
= .
(u − u
) du = 2
du = 2
− 2u
3u
3
u
1
1
1
√
7. (a) With u = 1 + e x one has du =
u = 1 + e2 . Thus,
Z
4
1
√
e x
√
dx =
√
x (1 + e x )
Z
1
√
·e
2 x
1+e2
1+e
√
x dx.
Now x = 1 gives u = 1 + e and x = 4 gives
1+e2
2
du = 2
u
1+e
ln u = 2 ln(1 + e2 ) − 2 ln(1 + e).
(b) A natural substitution is u = ex +1 leading to du = ex dx and so dx = du/ex = du/(u−1).
When x = 0 one has u = 2, and when x = 1/3 one has u = e1/3 + 1. Thus,
Z
1/3
0
1
dx =
x
e +1
Z
e1/3 +1
2
e1/3 +1
=
2
because
1
3
1
du =
u(u − 1)
(ln |u − 1| − ln |u|) =
1
3
Z
2
e1/3 +1 1
1
−
u−1 u
du
− ln(e1/3 + 1) + ln 2 = ln 2 − ln(e−1/3 + 1)
− ln(e1/3 + 1) = ln[e1/3 /(e1/3 + 1)] = − ln(e−1/3 + 1).18
(c) With z 4 = 2x − 1 one has 4z 3 dz = 2 dx. Also x = 8.5 gives z = 2 and x = 41 gives z = 3.
The integral becomes
Z
3
2
2z 3
dz = 2
z2 − z
Z
3
2
z2
dz = 2
z−1
Z
2
3
1
z+1+
dz
z−1
3
1 2
=2
2 z + z + ln(z − 1) = 7 + 2 ln 2
2
9.7 Infinite intervals of integration
3. (a) See the answer given in the book.
√
Integration by parts also
works in this case, with f (x) = ln x and g ′ (x) = 1/ x.
p
√
17
The substitution u = 1 + x also works.
e−x
18
, the suggested substitution t = e−x , with dt = −e−x dx works as well.
Rewriting the integrand as
1 + e−x
Verify that you get the same answer.
16
39
(b) Using a simplified notation and the result in Example 1(a), we have
Z
0
∞
1
x−
λ
2
λe
−λx
dx = −
1
= 2 +2
λ
∞
0
Z
1
x−
λ
∞
0
2
e
−λx
xe−λx dx −
2
λ
+
Z
Z
0
∞
∞
1
2 x−
λ
e−λx dx =
0
e−λx dx
1
2
2
1
+ 2− 2 = 2
2
λ
λ
λ
λ
where we have used both the result of Example 1(a) and part (a) in order to derive the
penultimate equality.
(c) Using the result of part (b) in order to derive the penultimate equality, we get
Z
0
∞
1
x−
λ
3
λe
−λx
Z ∞ 1 3 −λx
1 2 −λx
x−
3 x−
dx = −
e
+
e
dx
λ
λ
0
0
Z 1 2 −λx
3 ∞
3 1
2
1
1
x−
λe
dx = − 3 + · 2 = 3
=− 3 +
λ
λ 0
λ
λ
λ λ
λ
∞
5. (a) f ′ (x) = (1 − 3 ln x)/x4 = 0 at x = e1/3 , with f ′ (x) > 0 for x < e1/3 and f ′ (x) < 0 for
x > e1/3 . Hence f has a maximum at (e1/3 , 1/3e). Since f (x) → −∞ as x → 0+ , there
is no minimum. Use l’Hôpital’s rule to show that f (x) → 0 as x → ∞.
Rb
Rb
b
b
(b) Here, a x−3 ln x dx = − a 21 x−2 ln x + a 21 x−3 dx = a − 12 x−2 ln x − 14 x−2 .
R ∞ −3
This diverges when b = 1 and a → 0. But 1 x ln x dx = 1/4.
7. Provided that both limits exist, the integral is the sum of
Z 3
Z
√
I1 = lim
(1/ x + 2 ) dx and I2 = lim
ε→0+
ε→0+
−2+ε
−2
√
(1/ 3 − x ) dx.
√
√
√
3
x
+
2
)
=
lim
5
−
2
ε
)
=
2
5,
(2
+ (2
ε→0
−2+ε
√
√
√
√
3−ε
−2 (−2 3 − x ) = limε→0+ (−2 ε + 2 5 ) = 2 5.
√
Here, I1 = limε→0+
and I2 = limε→0+
3−ε
1
x−µ
12. All three parts (a)–(c) use the substitution u = √ , which gives du = √ dx, and so
2σ
σ 2
√
dx = σ 2 du.
Z +∞
Z +∞
1
2
e−u du = 1, by Eq. (9.7.8).
f (x) dx = √
(a)
π −∞
−∞
Z +∞
Z +∞
√
1
2
(µ + 2σu) e−u du = µ, using part (a) and Example 9.7.3.
xf (x) dx = √
(b)
π −∞
−∞
Z +∞
Z +∞
Z +∞
1
2
2
2 √
u2 e−u du.
(c) First,
(x − µ)2 f (x) dx =
2σ 2 u2 √ e−u σ 2 du = σ 2 √
π
σ 2π
−∞
−∞
−∞
R
R b 2 −u2
2
2
b
b
Now integrating by parts yields a u e
du = − 12 a ue−u + a 21 e−u du for all real
b
2
2
2
a, b. Because a ue−u = be−b − ae−a → 0 as a → −∞ and b → ∞, it follows that
R +∞ 2 −u2
R +∞
R +∞
√
2
du = −∞ 12 e−u du = 21 π by part (a). Hence −∞ (x − µ)2 f (x) dx = σ 2 .
−∞ u e
40
9.8 A glimpse at differential equations
x
10. (a) If f 6= r, the equation can be rewritten as ẋ = (r − f )x 1 −
.
(1 − f /r)K
There are two constant solutions x ≡ 0 and x ≡ (1−f /r)K, though the latter is negative,
so biologically meaningless, unless f ≤ r. With r̄ = r − f and K̄ = (1 − f /r)K, the
equation is x = r̄x(1 − x/K̄). Using (9.8.7), the solution is
x(t) =
K̄
(1 − f /r)K
=
(1 − f /r)K − x0 −(r−f )t
K̄ − x0 −r̄t
1+
e
e
1+
x0
x0
In the special case when f = r, the equation reduces to ẋ = −rx2 /K. Separating the
variables gives −dx/x2 = (r/K) dt, whereupon integration gives 1/x = (r/K)t + C. If
x(0) = x0 , we get C = 1/x0 , so the solution is x(t) = 1/[(r/K)t + 1/x0 ] → 0 as t → ∞.
(b) When f > r, the solution to the differential equation given in part (a) is still valid even
though both K̄ and r̄ are negative. Because the fish rate f exceeds the replenishment
rate r, however, the fish stock steadily declines. Indeed, as t → ∞ one has e−(r−f )t → ∞
and in fact the solution of the equation satisfies x(t) → 0. That is, the fish stock tends
to extinction.
9.9 Separable and linear differential equations
e2t dt. Integrate: 13 x3 = 12 e2t + C1 . Solve for x:
q
3 2t
3 2t
3
x = 2 e + 3C1 = 2 e + C, with C = 3C1 . Hence, x = 3 32 e2t + C.19
R
R
(b) dx/dt = e−t ex , so e−x dx = e−t dt. Integrate: −e−x = −e−t + C1 . Solve for x:
e−x = e−t + C, with C = −C1 . Hence, −x = ln(e−t + C), so x = − ln(e−t + C).
2. (a) dx/dt = e2t /x2 . Separate:
R
x2 dx =
R
(c) This follows directly from (9.9.3).
(d) Similar to (a).
R
R
(e) By Eq. (9.9.5), x = Ce2t + e2t (−t)e−2t dt = Ce2t − e2t te−2t dt. Here
Z
Z
1 −2t 1
1
1 −2t
−2t
−2t
te dt = t −
e dt = − t −
e
+
e
2
2
2
4
and thus
1
1
e−2t = Ce2t + t + .
2
4
R
R
2
2
2
(f) By (9.9.5), x = Ce−3t + e−3t e3t tet −3t dt = Ce−3t + e−3t tet dt = Ce−3t + 21 et −3t .
2t
x = Ce − e
2t
1
1
− t−
2
4
5. From (iii), L = L0 eβt , so K̇ = γK α L0 eβt , a separable equation. Writing the equation as
K −α K̇ = γL0 eβt , we get
Z
Z
1
γL0 βt
−α
K dK = γL0 eβt dt, so
K 1−α =
e + C1 .
1−α
β
19
1 3
x
3
It is important to include
q the constant at the integration step. Adding it only at the end would lead to an error:
= 21 e2t , x3 = 32 e2t , x =
3
3 2t
e
2
+ C. This is a solution only if C = 0, and is not the general solution.
41
(1 − α)γL0
(1 − α)γL0 βt
e + (1 − α)C1 , and at t = 0, K01−α =
+ (1 − α)C1 .
β
β
(1 − α)γL0 βt
It follows that K 1−α =
(e − 1) + K01−α , from which we find K.
β
Hence, K 1−α =
Review exercises for Chapter 9
4. (a) 5/4. (See Example 9.7.2.)
(b)
(c)
(d)
(e)
(f)
(g)
(h)
1 1
4 5
0 20 (1 + x ) = 31/20.
R
∞
−t − ∞ 5e−t dt = 5 ∞ e−t = −5.
0 5te
0
0
Re
Re
e
e
2
2
x(ln
x)
−
2
(ln
x)
dx
=
1
1
1 ln x dx = e − 2 1 (x ln x − x) = e − 2.
2 2 3
3/2 = 2 (93/2 − 1) = 52/9.
9
0 9 (x + 1)
0
1
1
1
3z
−∞ 3 ln(e + 5) = 3 (ln 6 − ln 5) = 3 ln(6/5).
R
1 e/2 4
1
1
1 e/2 3
4
4
4
4
4 1/2 x ln(2x) − 4 1/2 x dx = 4 (e/2) − 16 [(e/2) − (1/2) ] = (1/256)(3e
√
Introduce u = x. Then u2 = x, so 2u du = dx, and
Z
∞
1
√
e−
√
x
x
dx =
Z
∞
1
e−u 2u
du = 2
u
Z
∞
+ 1).
e−u du = 2e−1 .
1
√
5. (a) With u = 9 + x one has x = (u − 9)2 and so dx = 2(u − 9) du. Also, u = 9 when x = 0
and u = 14 when x = 25. Thus,
Z 25
Z 14
Z 14 1
2(u − 9)
18
14
√ dx =
2−
du =
du = 10 − 18 ln .
u
u
9
9+ x
0
9
9
√
(b) With u = t + 2 one has t = u2 − 2 and so dt = 2u du. Also, u = 2 when t = 2 and
u = 3 when t = 7. Hence,
Z
7
√
t t + 2 dt =
2
Z
3
2
3
3
1 5 2 3
u − u
(u − 2u ) du = 2
(u − 2)u · 2u du = 2
5
3
2
2
243 54
886
422 76
32 16
=2
−
−
−
=
.
−
=
5
3
5
3
5
3
15
Z
2
4
2
√
(c) With u = 3 19x3 + 8 one has u3 = 19x3 + 8 and so 3u2 du = 57x2 dx. Also, x = 0 gives
u = 2 and x = 1 gives u = 3. Therefore
Z 1
Z 3
3
p
3 4
3
u = 195/4
57x2 19x3 + 8 dx =
3u3 du =
0
2
2 4
10. (a) As in Example 9.4.3, first we need to find P ∗ and Q∗ . From the equilibrium condition
f (Q∗ ) = 100 − 0.05Q∗ = g(Q∗ ) = 0.1Q∗ + 10, we obtain 0.15Q∗ = 90, and so Q∗ = 600.
Then P ∗ = g(Q∗ ) = 0.1Q∗ + 10 = 70. Moreover,
Z 600
Z 600
600
(30 − 0.05Q) dQ =
(f (Q) − P ∗ ) dQ =
CS =
30Q − 12 0.05Q2 = 9000
0
0
0
Z 600
Z 600
600
60Q − 12 0.1Q2 = 18000
(60 − 0.1Q) dQ =
(P ∗ − g(Q)) dQ =
PS =
0
0
0
42
(b) Equilibrium occurs when 50/(Q∗ + 5) = 4.5 + 0.1Q∗ . Clearing fractions and then simplifying, we obtain (Q∗ )2 + 50Q∗ − 275 = 0. The only positive solution is Q∗ = 5, and then
P ∗ = 5.
Z 5
5
50
CS =
− 5 dQ = [50 ln(Q + 5) − 5Q] = 50 ln 2 − 25
Q+5
0
0
Z 5
5
(5 − 4.5 − 0.1Q) dQ = (0.5Q − 0.05Q2 ) = 2.5 − 1.25 = 1.25
PS =
0
0
11. (a) f ′ (t) = 4[2 ln t · (1/t) · t − (ln t)2 · 1]/t2 = 4(2 − ln t) ln t/t2 , and
[2 · (1/t) − 2 ln t · (1/t)] t2 − [2 ln t − (ln t)2 ] 2t
(ln t)2 − 3 ln t + 1
=
8
t4
t3
(b) Here, f ′ (t) = 0 ⇐⇒ ln t(2 − ln t) = 0 ⇐⇒ ln t = 2 or ln t = 0 ⇐⇒ t = e2 or t = 1.
But f ′′ (1) = 8 > 0 and f ′′ (e2 ) = −8e−6 , so t = 1 is a local minimum point
and t = e2 ≈ 7.4 is a local maximum point. We find f (1) = 0 and f (e2 ) = 16e−2 ≈ 2.2.
Z e2
1
(c) The function is positive on [1, e2 ], so the area is A = 4
(ln t)2 dt. With u = ln t as
t
1
1
a new variable, du = dt. When t = 1 then u = 0, and t = e2 implies u = 2.
t
Z 2
2
1 3 32
2
u = .
u du = 4
Hence, A = 4
3
0 3
0
R −2
R
13. (a) Separable.
x dx = t dt, and so −1/x = 12 t2 + C1 , or x = 1/(C − 21 t2 ) (with
C = −C1 ).
f ′′ (t) = 4
(b) Direct use of (9.9.3).
(c) Direct use of (9.9.3).
R
(d) Using (9.9.5), x = Ce−5t + 10e−5t te5t dt.
R 5t
R
1 5t
e .
Here te dt = t 51 e5t − 15 e5t dt = 15 te5t − 25
1 5t
1 5t
−5t
−5t
Thus x = Ce
+ 10e ( 5 te − 25 e ) = Ce−5t + 2t − 52 .
R
R
(e) x = Ce−t/2 + e−t/2 et/2 et dt = Ce−t/2 + e−t/2 e3t/2 dt = Ce−t/2 + e−t/2 23 e3t/2
= Ce−t/2 + 32 et .
Z
Z
2
2 3t
−3t
−3t 1 2 3t
−3t
−3t
3t
t e dt = Ce
+e
(f) x = Ce
+e
t e −
te dt
3
3
Z
1
2
2
2
1
1 3t 1
= Ce−3t + t2 − e−3t
te −
e3t dt = Ce−3t + t2 − t + .
3
3
3
3
3
9
27
10 Topics in Financial Mathematics
10.2 Continuous compounding
6. With g(x) = (1 + r/x)x for all x > 0 one has ln g(x) = x ln(1 + r/x). Differentiating gives
g ′ (x)/g(x) = ln(1 + r/x) + x(−r/x2 )/(1 + r/x) = ln(1 + r/x) − (r/x)/(1 + r/x),
as claimed in the problem. Putting h(u) = ln(1+u)−u/(1+u), one has h′ (u) = u/(1+u)2 > 0
for u > 0, so h(u) > 0 for all u > 0, implying that g ′ (x)/g(x) = h(r/x) > 0 for all x > 0. So
g(x) is strictly increasing for x > 0. Because g(x) → er as x → ∞, it follows that g(x) < er
for all x > 0. Continuous compounding of interest is best for the lender.
43
10.4 Geometric series
6. Let x denote the number of years beyond 1971 that the extractable resources of iron will
last. We need to solve the equation 794 + 794 · 1.05 + · · · + 794 · (1.05)x = 249 · 103 . Using
Eq. (10.4.3) to sum the left hand side, the equation is 794[(1.05)x+1 − 1]/(1.05 − 1) = 249 · 103
or (1.05)x+1 = 249 · 103 · 0.05/794 + 1 = 12450/794 + 1 ≈ 16.68. Using a calculator, we find
x ≈ (ln 16.68/ ln 1.05) − 1 ≈ 56.68, so the resources will be exhausted part way through the
year 2028.
8. (a) The quotient of this infinite series is e−rt , so the sum is f (t) =
P (t)e−rt
P (t)
= rt
.
−rt
1−e
e −1
P ′ (t)(ert − 1) − P (t)rert
. Now t∗ > 0 can maximize f (t)
(ert − 1)2
∗
∗
∗
only if f ′ (t∗ ) = 0, or P ′ (t∗ )(ert − 1) = rP (t∗ )ert , so P ′ (t∗ )/P (t∗ ) = r/(1 − e−rt ).
r
1
1
(c) lim
∗ = “0/0” = lim ∗ −rt∗ = ∗
−rt
r→0 1 − e
r→0 t e
t
P∞
P
p
11. See Fig. SM10.4.11. If p > 1, then
(1/np ) = 1 + ∞
n=1
n=2 (1/n ) is finite because
P∞
p
n=2 (1/n ) is the sum of the shaded rectangles, and this sum is certainly less than the
area under the curve y = 1/xp over [1, ∞), which is equal to 1/(p − 1).
P
p
If p ≤ 1, the sum ∞
n=1 (1/n ) is the sum of the larger rectangles in the figure, and this sum
is larger than the area under the curve y = 1/xp over [1, ∞), which is unbounded when p ≤ 1.
P
p
Hence, ∞
n=1 (1/n ) diverges in this case.
(b) By direct computation, f ′ (t) =
y
y = 1/x p
1
1
2
3
4
x
Figure SM10.4.11
10.7 Internal rate of return
5. According to (10.7.1), the internal rate must satisfy
−100 000 +
10 000
10 000
10 000
+
+ ··· +
=0
1+r
(1 + r)2
(1 + r)20
After dividing all the terms by 10 000, and putting s = 1/(1 + r), we have to show that
the equation f (s) = s20 + s19 + · · · + s2 + s − 10 = 0 has a unique positive solution. Since
f (0) = −10 and f (1) = 10, the intermediate value theorem (Theorem 7.10.1) implies that
there exists a number s∗ between 0 and 1 such that f (s∗ ) = 0. This s∗ is the unique positive
root because f ′ (s) > 0 for all s > 0. In fact, from (10.4.3), f (s) = −10 + (s − s21 )/(1 − s), and
f (s∗ ) = 0 ⇐⇒ (s∗ )21 −11s∗ +10 = 0. Exercise 7.R.26 asks for an approximation to the unique
root of this equation in the interval (0, 1). An answer is s∗ ≈ 0.928, so r∗ = 1/s∗ − 1 ≈ 0.0775,
which means that the internal rate of return is about 7 43 %.
44
Review exercises for Chapter 10
9. We use formula (10.5.3) on the future value of an annuity: (5000/0.04)[(1.04)4 −1] = 21 232.32.
10. The last of seven payments will be on 1st January 2016, when the initial balance of 10 000 will
have earned interest for 10 years. So K must solve 10 000 · (1.04)10 + K[(1.04)8 − 1]/0.04 =
70 000. We find that K ≈ 5990.49.
√
1
13. (a) f ′ (t) = 100e t/2 e−rt 4√
−
r
. We see that f ′ (t) = 0 for t = t∗ = 1/16r2 . Since f ′ (t) > 0
t
for t < t∗ and f ′ (t) < 0 for t > t∗ , it follows that t∗ maximizes f (t).
√
(b) f ′ (t) = 200e−1/t e−rt t12 − r . We see that f ′ (t) = 0 for t = t∗ = 1/ r. Since f ′ (t) > 0
for t < t∗ and f ′ (t) < 0 for t > t∗ , it follows that t∗ maximizes f (t).
11 Functions of Many Variables
11.2 Partial derivatives with two variables
5. Parts (a)–(c) are easy. As for (d), note that z = xy = (eln x )y = ey ln x = eu with u =
y ln x. Then zx′ = eu u′x = xy (y/x) = yxy−1 . Similarly zy′ = eu u′y = xy ln x. Moreover,
′′ = (∂/∂x)(yxy−1 ) = y(y − 1)xy−2 .20 Similarly z ′′ = (∂/∂y)(xy ln x) = xy (ln x)2 and
zxx
yy
′′ = (∂/∂y)(yxy−1 ) = xy−1 + yxy−1 ln x.
zxy
11.3 Geometric representation
9. (a) It helps to regard the figure as a contour map of a mountain, whose level curves join points
at the same height above mean sea level. Near P the terrain is rising in the direction
of the positive x-axis, so fx′ (P ) > 0, and it is also sloping down in the direction of the
positive y-axis so fy′ (P ) < 0. Near Q, the terrain slopes in the opposite directions. Hence
fx′ (Q) < 0 and fy′ (Q) > 0.
(b) (i) The line x = 1 has no point in common with any of the given level curves.
(ii) The line y = 2 intersects the level curve z = 2 at x = 2 and x = 6 (approximately).
(c) If you start at the point (6, 0) and move up along the line 2x + 3y = 12, the first marked
level curve you meet is z = f (x, y) = 1. Moving further you meet level curves with higher
z-values. The level curve with the highest z-value you meet is z = 3, which is the level
curve that just touches the straight line.
R1
10. The inequalities on the partial derivatives imply that: F (1, 0) − F (0, 0) = 0 F1′ (x, 0) dx ≥
R1
R2 ′
R1 ′
0 2 dx = 2; F (2, 0)R − F (1, 0) = 1 F1 (x, 0) dx ≥ 2; F (0, 1) − RF (0, 0) = 0 F2 (0, y) dy ≤ 1;
1
1
F (1, 1) − F (0, 1) = 0 F1′ (x, 1) dx ≥ 2; and F (1, 1) − F (1, 0) = 0 F2′ (1, y) dy ≤ 1.
11.5 Functions of more variables
3. (a) In the first week it buys 120/50 = 2.4 million shares, followed successively by 120/60 =
2 million shares, 120/45 = 2.667 million, 120/40 = 3 million, 120/75 = 1.6 million, and
finally 120/80 = 1.5 million in the sixth week. The total is 13.167 million shares.
20
When differentiating xy−1 partially w.r.t. x, one treats y as a constant, so the rule dxa /dx = axa−1 applies.
45
(b) The arithmetic mean price in dollars is 350/6 = 58.33. But the total cost at that price of
the 13.167 million shares which the fund has acquired would be 13.167 × 58.33 = 768.031
million dollars. So using the arithmetic mean price would overstate the fund’s cost by
48.031 million dollars. A more accurate statement of the mean price is 720 / 13.167 =
54.68 dollars per share. Routine arithmetic shows that this is the harmonic mean of the
six prices defined by formula (c) of Example 2.
11.6 Partial derivatives with more variables
2. Parts (a)–(d) are routine. As for (e) and (f):
(e) f (x, y, z) = (x2 + y 3 + z 4 )6 = u6 , with u = x2 + y 3 + z 4 . Then:
f1′ = 6u5 u′1 = 6(x2 + y 3 + z 4 )5 2x = 12x(x2 + y 3 + z 4 )5
f2′ = 6u5 u′2 = 6(x2 + y 3 + z 4 )5 3y 2 = 18y 2 (x2 + y 3 + z 4 )5
f3′ = 6u5 u′3 = 6(x2 + y 3 + z 4 )5 4z 3 = 24z 3 (x2 + y 3 + z 4 )5
(f) f (x, y, z) = exyz = eu , with u = xyz, gives f1′ = eu u′1 = exyz yz.
Similarly, f2′ = eu u′2 = exyz xz, and f3′ = eu u′3 = exyz xy.
z
10. From f = xy we get ln f = y z ln x.
z
z
Differentiating w.r.t. x yields fx′ /f = y z /x, and so fx′ = f y z /x = xy y z /x = y z xy −1 .
z
Differentiating w.r.t. y yields fy′ /f = zy z−1 ln x, and so fy′ = zy z−1 (ln x)xy .
z
Differentiating w.r.t. z yields fz′ /f = y z (ln y)(ln x), and so fz′ = y z (ln x)(ln y)xy .
11. For (x, y) 6= (0, 0) one has
f1′ (x, y) = y(x4 + 4x2 y 2 − y 4 )(x2 + y 2 )−2
and
f2′ (x, y) = x(x4 − 4x2 y 2 − y 4 )(x2 + y 2 )−2
Thus, for y 6= 0, f1′ (0, y) = −y. Because f1′ (0, 0) = limh→0 [f (h, 0) − f (0, 0)]/h = 0, this is
also correct for y = 0. Similarly, f2′ (x, 0) = x for all x.
′′ (0, y) = (∂/∂y)f ′ (0, y) = −1 for all y. In particular, f ′′ (0, 0) = −1.
It follows that f12
1
12
′′ (x, 0) = (∂/∂x)f ′ (x, 0) = 1 for all x, and so f ′′ (0, 0) = 1.
Similarly, f21
2
21
Straightforward differentiation shows that, for (x, y) 6= (0, 0),
′′
′′
f12
(x, y) = f21
(x, y) =
x6 + 9x4 y 2 − 9x2 y 4 − y 6
.
(x2 + y 2 )3
(∗)
Thus, away from the origin, the two cross partial derivatives are equal, in accordance with
′′ (0, 0) = −1 and f ′′ (0, 0) = 1.
Young’s theorem. At the origin, however, we have seen that f12
21
′′
′′
Therefore, at least one of f12 and f21 must be discontinuous there. Indeed, it follows from (∗)
′′ (x, 0) = 1 for all x 6= 0 and f ′′ (0, y) = −1 for all y 6= 0. Thus, as close to (0, 0) as we
that f12
12
′′ equals 1 and also points where f ′′ equals −1. Therefore
want, we can find points where f12
12
′′ cannot be continuous at (0, 0). The same argument shows that f ′′ cannot be continuous
f12
21
at (0, 0).
46
11.7 Economic applications
2. (a) YK′ = aAK a−1 and YK′ = aBLa−1 , so KYK′ +LYL′ = aAK a +aBLa = a(AK a +BLa ) = aY.
(b) KYK′ + LYL′ = KaAK a−1 Lb + LAK a bLb−1 = aAK a Lb + bAK a Lb = (a + b)Y .
(c) YK′ =
2aKL5 − bK 4 L2
2bK 5 L − aK 2 L4
′
and
Y
=
, so KYK′ + LYL′ is
L
(aL3 + bK 3 )2
(aL3 + bK 3 )2
2aK 2 L5 − bK 5 L2 + 2bK 5 L2 − aK 2 L5
K 2 L2 (aL3 + bK 3 )
K 2 L2
=
=
= Y.
(aL3 + bK 3 )2
(aL3 + bK 3 )2
aL3 + bK 3
Note that according to Section 12.6 the functions in (a), (b), and (c) are homogeneous of
degrees a, a + b, and 1, respectively, so these results are immediate implications of Euler’s
Theorem, (12.6.2).
7. Here
YK′
YL′
=
=
=
=
−(µ/̺)−1
(−µ/̺)a(−̺)K −̺−1 Aeλt [aK −̺ + bL−̺ ]
−(µ/̺)−1
µaK −̺−1 Aeλt [aK −̺ + bL−̺ ]
,
−(µ/̺)−1
−̺−1
λt
−̺
−̺
(−µ/̺)b(−̺)L
Ae [aK + bL ]
−(µ/̺)−1
µbL−̺−1 Aeλt [aK −̺ + bL−̺ ]
−(µ/̺)−1
= µY .
Thus, KYK′ + LYL′ = µ(aK −̺ + bL−̺ )Aeλt [aK −̺ + bL−̺ ]
Once again, this function is homogeneous of degree µ, so the result is an immediate implication
of Euler’s Theorem, (12.6.2).
11.8 Partial elasticities
′ −D
mDm
p
pD
pD
∂
′
=p
= 2 (mDm
− D) = 2 (Elm D − 1) > 0 if and only if
4. Note that
2
∂m m
m
m
m
Elm D > 1. So pD/m increases with m if Elm D > 1.21
Review exercises for Chapter 11
12. (a) See the answer given in the book.
(b) We want to find all (x, y) that satisfy: (i) 4x3 − 8xy = 0 and (ii) 4y − 4x2 + 4 = 0. From
equation (i), 4x(x2 − 2y) = 0, which implies that x = 0 or x2 = 2y.
For x = 0, equation (ii) yields y = −1, so (x, y) = (0, −1) is one solution.
√
For x2 = 2y, (ii) reduces to 4y − 8y + 4 = 0, or y = 1. But then x2 = 2, so x = ± 2.
√
Hence, two additional solutions are (x, y) = (± 2, 1).
12 Tools for Comparative Statics
12.1 A simple chain rule
6. (a) If z = F (x, y) = x + y with x = f (t) and y = g(t), then F1′ = F2′ = 1, so the chain rule,
formula (12.1.1), gives dz/dt = 1 · f ′ (t) + 1 · g ′ (t) = f ′ (t) + g ′ (t).
21
Using the formulas in Exercise 7.7.9, the result also follows from the fact that Elm (pD/m) = Elm p + Elm D −
Elm m = Elm D − 1.
47
(b) This case is like (a), except that F2′ = −1 so the chain rule gives dz/dt = f ′ (t) − g ′ (t).
(c) If z = F (x, y) = xy with x = f (t) and y = g(t), then F1′ (x, y) = y, F2′ (x, y) = x,
dx/dt = f ′ (t), and dy/dt = g ′ (t), so formula (12.1.1) gives
dx
dy
dz
= F1′ (x, y)
+ F2′ (x, y)
= yf ′ (t) + xg ′ (t) = f ′ (t)g(t) + f (t)g ′ (t).
dt
dt
dt
x
1
x
with x = f (t) and y = g(t), then F1′ (x, y) = , F2′ (x, y) = − 2 ,
y
y
y
dx
dy
= f ′ (t), and
= g ′ (t). So formula (12.1.1) gives
dt
dt
(d) If z = F (x, y) =
dx
dy
1
x
yf ′ (t) − xg ′ (t)
f ′ (t)g(t) − f (t)g ′ (t)
dz
= F1′ (x, y) +F2′ (x, y)
= f ′ (t)− 2 g ′ (t) =
=
.
dt
dt
dt
y
y
y2
(g(t))2
(e) If z = F (x, y) = G(x), independent of y, with x = f (t), then F1′ = G′ and F2′ = 0, so
(12.1.1) gives dz/dt = G′ (x) · f ′ (t), which is the chain rule for one variable.
7. Let U (x) = u(x, h(x)). Then
′
U (x) =
u′1
+
u′2 h′ (x)
αxα−1
+
= α
x + zα
α
αz α−1
−
α
α
x +z
z
4a 3
x (ax4 + b)−2/3
3
Because the term in large parentheses equals −αxα /z(xα + z α ), simplifying gives
U ′ (x) =
4ax3
αxα
αxα−1 (3z 3 − 4ax4 )
αxα−1
−
=
xα + z α z(xα + z α ) 3z 2
3(xα + z α )z 3
But z 3 = ax4 + b so 3z 3 − 4ax4 = 3b − ax4 . It follows that U ′ (x) = 0 when x = x∗ =
whereas U ′ (x) > 0 for x < x∗ and U ′ (x) < 0 for x > x∗ . Hence x∗ maximizes U .
p
4
3b/a,
2
2
′
′
8. Differentiating
(12.1.1)
2 (x, y) dy/dt].
w.r.t.
t yields d z/dt = (d/dt)[F
1 (x, y) dx/dt]+(d/dt)[F
d
dy dx
d2 x
dx
dx
′′
′′
Here,
+ F12
(x, y)
+ F1′ (x, y) 2 ,
F1′ (x, y)
= F11
(x, y)
dt
dt
dt
dt dt
dt
d
dy dy
d2 y
dy
dx
′′
′′
while
+ F22
(x, y)
+ F2′ (x, y) 2 .
F2′ (x, y)
= F21
(x, y)
dt
dt
dt
dt dt
dt
′′ = F ′′ .
The conclusion follows from summing these two while assuming that F12
21
12.2 Chain rules for many variables
2. (a) Let z = F (x, y) = xy 2 with x = t + s2 and y = t2 s. Then F1′ (x, y) = y 2 , F2′ (x, y) = 2xy,
∂x/∂t = 1, and ∂y/∂t = 2ts. Now (12.2.1) gives
∂z
∂x
∂y
= F1′ (x, y)
+ F2′ (x, y)
= y 2 + 2xy2ts = (t2 s)2 + 2(t + s2 )t2 s2ts = t3 s2 (5t + 4s2 ).
∂t
∂t
∂t
The other partial derivative ∂z/∂s is found in the same way.
2y
∂x
∂y
−2sx ts
∂z
= F1′ (x, y)
+ F2′ (x, y)
=
et+s +
e , etc.
(b)
2
∂t
∂t
∂t
(x + y)
(x + y)2
48
12. (a) Let v = x3 +y 3 +z 3 −3xyz, so that u = ln v. Then ∂u/∂x = (1/v)(∂v/∂x) = (3x2 −3yz)/v.
Similarly, ∂u/∂y = (3y 2 − 3xz)/v, and ∂u/∂z = (3z 2 − 3xy)/v. Hence,
x
∂u
∂u
∂u
1
1
1
3v
+y
+z
= (3x3 − 3xyz) + (3y 3 − 3xyz) + (3z 3 − 3xyz) =
=3
∂x
∂y
∂z
v
v
v
v
which proves (i). Equation (ii) is then proved by elementary algebra.
(b) Note that f is here a function of one variable.
With z = f (u) where u = x2 y, we get ∂z/∂x = f ′ (u)u′x = 2xyf ′ (x2 y).
Likewise, ∂z/∂y = x2 f ′ (x2 y), so x ∂z/∂x = 2x2 yf ′ (x2 y) = 2y ∂z/∂y.
12.3 Implicit differentiation along a level curve
2. (a) See the answer given in the book.
′′ = 0, F ′′ = 3, and
(b) Let F (x, y) = x − y + 3xy. Then F1′ = 1 + 3y, F2′ = −1 + 3x, F11
12
′′ = 0. So y ′ = −F ′ /F ′ = −(1 + 3y)/(−1 + 3x). Moreover, using Eq. (12.3.4),
F22
1
2
y ′′ = −
6(1 + 3y)(−1 + 3x)
6(1 + 3y)
1 ′′ ′ 2
′′ ′ ′
′′
(F1′ )2 =
F11 (F2 ) − 2F12
F1 F2 + F22
=
.
′
3
3
(F2 )
(−1 + 3x)
(−1 + 3x)2
′′ = −30x4 , F ′′ = 0, F ′′ = 20y 3 ,
(c) Let F (x, y) = y 5 − x6 . Then F1′ = −6x5 , F2′ = 5y 4 , F11
12
22
so y ′ = −F1′ /F2′ = −(−6x5 /5y 4 ) = 6x5 /5y 4 . Moreover, using Eq. (12.3.4),
y ′′ = −
6x4 144x10
1 4
4 2
3
5 2
= 4 −
(−30x
)(5y
)
+
20y
(−6x
)
.
(5y 4 )3
y
25y 9
3. (a) With F (x, y) = 2x2 + xy + y 2 , one has y ′ = −F1′ /F2′ = −(4x + y)/(x + 2y) = −4 at (2, 0).
Moreover, using (12.3.4) gives y ′′ = −(28x2 + 14y 2 + 14xy)/(x + 2y)3 . At (2, 0) this gives
y ′′ = −14. The point–slope formula for the tangent gives y = −4x + 8.
(b) y ′ = 0 requires y = −4x. Inserting this into the original equation gives a quadratic
√
equation 14x2 − 8 = 0 for x. The roots of this equation are x = ±2 7/7. Along with the
corresponding values of y, this gives the two points indicated in the text.
12.4 More general cases
√
3. (a) Here, Eq. (∗) is P/2 L∗ = w. Solve for L∗ . The rest is routine.
(b) The first-order condition is now
P f ′ (L∗ ) − CL′ (L∗ , w) = 0
(∗)
To find the partial derivatives of L∗ , we will differentiate (∗) partially w.r.t. P and w.
First, we find the partial derivative of P f ′ (L∗ ) w.r.t. P using the product rule. The
result is f ′ (L∗ ) + P f ′′ (L∗ )(∂L∗ /∂P ). Then the partial derivative of CL′ (L∗ , w) w.r.t. P is
′′ (L∗ , w)(∂L∗ /∂P ). So differentiating (∗) w.r.t. P yields
CLL
f ′ (L∗ ) + P f ′′ (L∗ )
∂L∗
∂L∗
′′
− CLL
(L∗ , w)
=0
∂P
∂P
Solving for ∂L∗ /∂P gives the answer.
49
Second, differentiating (∗) w.r.t. w yields
′′
[P f ′′ (L∗ ) − CLL
(L∗ , w)]
∂L∗
′′
− CLw
(L∗ , w) = 0.
∂w
Solving for ∂L∗ /∂w gives the answer.
6. F1′ (x, y) = ey−3 + y 2 and F2′ (x, y) = xey−3 + 2xy − 2. Hence, the slope of the tangent to the
level curve F (x, y) = 4 at the point (1, 3) is y ′ = −F1′ (1, 3)/F2′ (1, 3) = −10/5 = −2. Then
use the point–slope formula.
7. Taking the logarithm of both sides, we get (1 + c ln y) ln y = ln A + α ln K + β ln L.
c ∂y
1 ∂y
α
Differentiating partially with respect to K gives
ln y + (1 + c ln y)
= .
y ∂K
y ∂K
K
Solving for ∂y/∂K yields the given answer. Then ∂y/∂L is found in the same way.
12.5 Elasticity of substitution
3. When F (K, L) = AK a Lb , the first- and second-order partial derivatives satisfy
′ = aF/K, F ′ = bF/L, F ′′
2
′′
′′
2
FK
L
KK = a(a − 1)F/K , FKL = abF/KL, and FLL = b(b − 1)F/L .
But then the numerator of the expression for σyx is
′
′
+ LFL′ ) = −(aF/K)(bF/L)(a + b)F = −ab(a + b)F 3 /KL,
−FK
FL′ (KFK
whereas the denominator is
′′
′
′′
′ 2 ′′
− 2FK
FL′ FKL
+ (FK
) FLL
KL (FL′ )2 FKK
= KLF 3 [b2 a(a − 1) − 2a2 b2 + a2 b(b − 1)]/K 2 L2 = −ab(a + b)F 3 /KL.
It follows that σKL = 1.
12.6 Homogeneous functions of two variables
3. That f is homogeneous of degree 3 is straightforward. As for the properties,
(12.6.2): xf1′ (x, y) + yf2′ (x, y) = x(y 2 + 3x2 ) + y(2xy) = 3(x3 + xy 2 ) = 3f (x, y);
(12.6.3): It is easy to see that f1′ (x, y) = y 2 + 3x2 and f2′ (x, y) = 2xy are homogeneous of
degree 2;
(12.6.4): f (x, y) = x3 + xy 2 = x3 [1 + (y/x)2 ] = x3 f (1, y/x) = y 3 [(x/y)3 + x/y] = y 3 f (x/y, 1);
′′ + 2xyf ′′ + y 2 f ′′ = x2 (6x) + 2xy(2y) + y 2 (2x) = 6x3 + 6xy 2 = 3 · 2f (x, y).
(12.6.5): x2 f11
12
22
9. Let C and D denote the the numerator and the denominator in the expression for σyx in
Exercise 12.5.3. Because F is homogeneous of degree one, Euler’s theorem implies that
′′ = −yF ′′ and yF ′′ = −xF ′′ = −xF ′′ . Hence,
C = −F1′ F2′ F , and (12.6.6) implies that xF11
12
22
21
12
2 ′ 2
′′
′′
′′
′′
D = xy (F2′ )2 F11
− 2F1′ F2′ F12
+ (F1′ )2 F22
= −F12
y (F2 ) + 2xyF1′ F2′ + x2 (F1′ )2
′′ 2
′′
F ,
(xF1′ + yF2′ )2 = −F12
= −F12
′′ F 2 ) = F ′ F ′ /F F ′′ .
using Euler’s theorem again. Hence σxy = C/D = (−F1′ F2′ F )/(−F12
1 2
12
50
12.7 Homogeneous and homothetic functions
1. Parts (a) and (f) are easy. Here are solutions of the other parts:
(b) Note that xgx′ + ygy′ + zgz′ = g(x, y, z) + 2. This does not equal kg(x, y, z) for any k, so
Euler’s theorem implies that g is not homogeneous of any degree.
√ √
√
√
√
√
√
t ( x + y + z)
tx + ty + tz
(c) h(tx, ty, tz) =
=
= t−1/2 h(x, y, z) for all t > 0,
tx + ty + tz
t(x + y + z)
so h is homogeneous of degree −1/2.
√
(tx)2 + (ty)2
t2 (x2 + y 2 )
√
(d) G(tx, ty) = txty ln
= t xy ln
= t G(x, y) for all t > 0, so G
txty
t2 xy
is homogeneous of degree 1.
(e) xHx′ +yHy′ = x(1/x)+y(1/y) = 2. Since 2 does not equal k(ln x+ln y) for any constant k,
Euler’s theorem implies that H is not homogeneous of any degree.
2. (a) We find that
1
(tx1 tx2 tx3 )2
1
1
f (tx1 , tx2 , tx3 ) =
+
+
(tx1 )4 + (tx2 )4 + (tx3 )4 tx1 tx2 tx3
1
1
1
1
t6 (x1 x2 x3 )2
= tf (x1 , x2 , x3 ),
+
+
· ·
= 4 4
x1 x2 x3
t (x1 + x42 + x43 ) t
so f is homogeneous of degree 1.
(b) We find that x is homogeneous of degree µ because
−µ/̺
x(tv1 , tv2 , . . . , tvn ) = A δ1 (tv1 )−̺ + δ2 (tv2 )−̺ + · · · + δn (tvn )−̺
h
i−µ/̺
= A t−̺ (δ1 v1−̺ + δ2 v2−̺ + · · · + δn vn−̺ )
i−µ/̺
h
= (t−̺ )−µ/̺ A δ1 v1−̺ + δ2 v2−̺ + · · · + δn vn−̺
h
i−µ/̺
= tµ x(x1 , x2 , x3 )
= tµ A δ1 v1−̺ + δ2 v2−̺ + · · · + δn vn−̺
5. (a) We use definition (12.7.6). Suppose (x1 y1 )2 + 1 = (x2 y2 )2 + 1. Then (x1 y1 )2 = (x2 y2 )2 .
If t > 0, then (tx1 ty1 )2 + 1 = (tx2 ty2 )2 + 1 ⇐⇒ t4 (x1 y1 )2 + 1 = t4 (x2 y2 )2 + 1. This
holds if and only if t4 (x1 y1 )2 = t4 (x2 y2 )2 ⇐⇒ (x1 y1 )2 = (x2 y2 )2 , so f is homothetic.
2(x1 y1 )2
2(x2 y2 )2
=
we get 2(x1 y1 )2 [(x2 y2 )2 + 1] = 2(x2 y2 )2 [(x1 y1 )2 + 1] and
(x1 y1 )2 + 1
(x2 y2 )2 + 1
so (x1 y1 )2 = (x2 y2 )2 . For all t > 0 one has
(b) From
2(tx1 ty1 )2
2(tx2 ty2 )2
2t4 (x1 y1 )2
2t4 (x2 y2 )2
=
⇐⇒
=
(tx1 ty1 )2 + 1
(tx2 ty2 )2 + 1
t4 (x1 y1 )2 + 1
t4 (x2 y2 )2 + 1
2
(x1 y1 )
(x2 y2 )2
⇐⇒ 4
=
t (x1 y1 )2 + 1
t4 (x2 y2 )2 + 1
Now for all a, b > 0 notice that
b
a
= 4
⇐⇒ abt4 + a = abt4 + b ⇐⇒ a = b.
t4 a + 1
t b+1
51
So we have proved that for all t > 0 one has
2(tx1 ty1 )2
2(tx2 ty2 )2
=
⇐⇒ (x1 y1 )2 = (x2 y2 )2 .
(tx1 ty1 )2 + 1
(tx2 ty2 )2 + 1
It follows that f is homothetic.
(c) f (1, 0) = 1 = f (0, 1), but f (2, 0) = 4 6= 8 = f (0, 2). This is enough to show that f is not
homothetic.
(d) g(x, y) = x2 y is homogeneous of degree 3 and u → eu is strictly increasing, so f is
homothetic according to (12.7.7).
7. Define ∆ = ln C(tw, y) − ln C(w, y). It suffices to prove that ∆ = ln t, because then
C(tw, y)/C(w, y) = e∆ = t. We find that
∆=
n
X
i=1
n
n
X
1 X
aij [ln(twi ) ln(twj ) − ln wi ln wj ] + ln y
ai [ln(twi ) − ln wi ] +
bi [ln(twi ) − ln wi ].
2
i,j=1
i=1
Since ln(twi ) − ln wi = ln t + ln wi − ln wi = ln t
and ln(twi ) ln(twj ) − ln wi ln wj = (ln t)2 + ln t ln wi + ln t ln wj , this reduces to
∆ = ln t
n
X
i=1
n
n
n
n
n
n
X
X
X
X
X
X
1
1
1
aij + ln t
ln wi
ln wj
aij + ln t
aij +ln y ln t
bi .
ai + (ln t)2
2
2
2
i,j=1
j=1
i=1
i=1
j=1
i=1
The restrictions on the parameters ai , aij , and bi imply that ∆ = ln t + 0 + 0 + 0 + 0 = ln t.
12.8 Linear approximations
7. We use formula (12.8.3).
(a) Here, ∂z/∂x = 2x and ∂z/∂y = 2y. At the point (1, 2, 5) we get ∂z/∂y = 2 and
∂z/∂x = 4, so the tangent plane has the equation z − 5 = 2(x − 1) + 4(y − 2) or
z = 2x + 4y − 5.
(b) From z = (y − x2 )(y − 2x2 ) = y 2 − 3x2 y + 2x4 we get ∂z/∂x = −6xy + 8x3 and ∂z/∂y =
2y − 3x2 . Thus, at the point (1, 3, 2) we have ∂z/∂x = −10 and ∂z/∂y = 3. The tangent
plane is given by the equation z − 2 = −10(x − 1) + 3(y − 3) or z = −10x + 3y + 3.
8. g(0) = f (x0 ), g(1) = f (x). Using formula (12.2.3), one derives
g ′ (t) = f1′ (x0 + t(x − x0 ))(x1 − x01 ) + · · · + fn′ (x0 + t(x − x0 ))(xn − x0n )
Putting t = 0 gives g ′ (0) = f1′ (x0 )(x1 − x01 ) + · · · + fn′ (x0 )(xn − x0n ), and the conclusion follows.
12.9 Differentials
4. T (x, y, z) = [x2 + y 2 + z 2 ]1/2 = u1/2 , where u = x2 + y 2 + z 2 . Then dT = 21 u−1/2 du =
u−1/2 (x dx + y dy + z dz). For x = 2, y = 3, and z = 6, we have u = 49, T = 7 and
dT = 71 (x dx + y dy + z dz) = 71 (2 dx + 3 dy + 6 dz). Thus,
T (2 + 0.01, 3 − 0.01, 6 + 0.02) ≈ T (2, 3, 6) +
52
1
[2 · 0.01 + 3(−0.01) + 6 · 0.02]
7
1
= 7 + · 0.11 ≈ 7.015714
7
A calculator gives the better approximation T (2.01, 2.99, 6.02) ≈
√
49.2206 ≈ 7.015739.
12.11 Differentiating systems of equations
3. Since we are asked to find the partials of y1 and y2 w.r.t. x1 only, we might as well differentiate
the system partially w.r.t. x1 only to obtain the two simultaneous equations:
3−
∂y1
∂y2
− 9y22
=0
∂x1
∂x1
3x21 + 6y12
and
∂y1
∂y2
−
=0
∂x1 ∂x1
Solving these for the partials gives the answers in the book.
An alternative method, in particular if one needs all the partials, is to use total differentiation:
3 dx1 + 2x2 dx2 − dy1 − 9y22 dy2 = 0
and
3x21 dx1 − 2 dx2 + 6y12 dy1 − dy2 = 0
Letting dx2 = 0 and solving for dy1 and dy2 leads to dy1 = A dx1 and dy2 = B dx1 , where
A = ∂y1 /∂x1 and B = ∂y2 /∂x1 .
4. Differentiating with respect to M , while remembering that Y and r are functions of the
′ = S ′ (Y )Y ′ and aY ′ + L′ (r)r ′ = 1.
independent variables a and M , one obtains I ′ (r)rM
M
M
M
Writing this as a linear equation system in standard form, we get
′
′
−S ′ (Y ) YM
+ I ′ (r) rM
=0
and
′
′
a YM
+ L′ (r) rM
=1
Using either ordinary elimination or formula (2.4.2) gives the solution
′
=
YM
I ′ (r)
S ′ (Y )L′ (r) + aI ′ (r)
and
′
=
rM
S ′ (Y )
S ′ (Y )L′ (r) + aI ′ (r)
Review exercises for Chapter 12
4. X = N g(u) where u = ϕ(N )/N . Then du/dN = [ϕ′ (N )N − ϕ(N )]/N 2 = (1/N )(ϕ′ (N ) − u).
Using the product rule and the chain rule for differentiation, one obtains
dX
du
= g(u) + N g ′ (u)
= g(u) + g ′ (u)(ϕ′ (N ) − u).
dN
dN
Differentiating the right-hand side w.r.t. N gives
d2 X
du ′
du
du
′
′′
′
′′
= g (u)
+ g (u)
ϕ (N ) − u + g (u) ϕ (N ) −
dN 2
dN
dN
dN
2
ϕ(N )
ϕ(N )
ϕ(N )
1
ϕ′ (N ) −
ϕ′′ (N )
+ g′
= g ′′
N
N
N
N
11. Taking the elasticity of each side of the equation gives Elx (y 2 ex e1/y ) = Elx y 2 + Elx ex +
Elx e1/y = 0. Here Elx y 2 = 2 Elx y and Elx ex = x. Moreover, Elx e1/y = Elx eu , where
u = 1/y, so Elx eu = u Elx (1/y) = (1/y)(Elx 1 − Elx y) = −(1/y) Elx y. All in all, 2 Elx y + x −
(1/y) Elx y = 0, so Elx y = xy/(1 − 2y).22
22
We used the rules for elasticities in Exercise 7.7.9. If you are not comfortable with these rules, you can find y ′
by implicit differentiation and then use the formula Elx y = (x/y)y ′ .
53
16. (a) Differentiating, then gathering all terms in dp and dL on the left-hand side, one has
(i) F ′ (L) dp + pF ′′ (L) dL = dw
(ii) F (L) dp + (pF ′ (L) − w) dL = L dw + dB
Since we know that pF ′ (L) = w, (ii) implies that dp = (Ldw + dB)/F (L). Substituting
into (i) and solving for dL, we obtain dL = [(F (L)−LF ′ (L))dw−F ′ (L)dB]/pF (L)F ′′ (L).
It follows that
∂p
L
=
,
∂w
F (L)
∂p
1
=
,
∂B
F (L)
∂L
F (L) − LF ′ (L)
=
,
∂w
pF (L)F ′′ (L)
∂L
F ′ (L)
=−
∂B
pF (L)F ′′ (L)
(b) Because all variables including p are positive, whereas F ′ (L) > 0 and F ′′ (L) < 0, it is
clear that ∂p/∂w, ∂p/∂B, and ∂L/∂B are all positive.
The sign of ∂L/∂w is the opposite of the sign of F (L) − LF ′ (L). From the equations in
the model, we get F ′ (L) = w/p and F (L) = (wL + B)/p, so F (L) − LF ′ (L) = B/p > 0.
Therefore ∂L/∂w < 0.
19. (a) The first-order condition for a maximum is P ′ (t) = V ′ (t)e−rt − rV (t)e−rt − me−rt = 0.
Cancelling e−rt , we see that t∗ can only maximize the present value provided (∗) is
satisfied. The equation says that the marginal increase V ′ (t∗ ) in market value per unit of
time from keeping the car a little longer must equal the sum of the interest cost rV (t∗ )
per unit time from waiting to receive the sales revenue, plus the maintenance cost m per
unit of time.
(b) Differentiating P (t) again gives
P ′′ (t) = V ′′ (t)e−rt − rV ′ (t)e−rt − rV ′ (t)e−rt + r2 V (t)e−rt + rme−rt .
Gathering terms, we have
P ′′ (t) = [V ′′ (t) − rV ′ (t)]e−rt + [V ′ (t) − rV (t) − m](−re−rt ).
At the critical point t∗ the last square bracket is 0, so the condition P ′′ (t∗ ) < 0 reduces
to D = V ′′ (t∗ ) − rV ′ (t∗ ) < 0.
(c) Taking the differential of (∗) yields V ′′ (t∗ ) dt∗ = dr V (t∗ ) + r V ′ (t∗ ) dt∗ = dm. Hence
∂t∗
V (t∗ )
V (t∗ )
= ′′ ∗
=
∂r
V (t ) − rV ′ (t∗ )
D
and
∂t∗
1
1
= ′′ ∗
=
′
∗
∂m
V (t ) − rV (t )
D
Assuming that V (t∗ ) > 0 (otherwise it would be better to scrap the car immediately),
both partial derivatives are negative. A small increase in either the interest rate or the
maintenance cost makes the owner want to sell the car a bit sooner.
13 Multivariable Optimization
13.2 Two variables: sufficient conditions
3. Solving the budget equation to express x as a function of y and z yields x = 108 − 3y − 4z.
Then utility as a function of y and z is U = (108−3y −4z)yz. Necessary first-order conditions
54
for a maximum are Uy′ = 108z − 6yz − 4z 2 = 0 and Uz′ = 108y − 3y 2 − 8yz = 0. Because y and
z are assumed to be positive, these two equations reduce to 6y + 4z = 108 and 3y + 8z = 108,
with solution y = 12 and z = 9.23
13.3 Local extreme points
3. (a) The first-order partial derivatives of f are f1′ = (2x − ay)ey , f2′ = x(x − ay − a)ey .
The second-order partial derivatives are
′′
f11
= 2ey ,
′′
f12
= (2x − ay − a)ey ,
′′
= x(x − ay − 2a)ey .
f22
The critical points where f1′ = 0 and f2′ = 0 are the solutions of the two-equation system:
(i) 2x − ay = 0;
(ii) x(x − ay − a) = 0
If x = 0, then (i) gives y = 0, because a 6= 0.
If x 6= 0, then (ii) gives x = ay + a, whereas (i) gives x = 21 ay. Hence x = −a and y = −2.
So there are two critical points, (0, 0) and (−a, −2).
To determine the nature of each critical point (x0 , y0 ), we use the second-derivative test,
′′ (x , y ), B = f ′′ (x , y ), and C = f ′′ (x , y ). The test gives:
with A = f11
0 0
12 0 0
22 0 0
Point
(0, 0)
(−a, −2)
A
2
2e
−2
B
C
−a
−ae
0
2 −2
a e
−2
AC − B 2
−a
2
2 −4
a e
Result
Saddle point
Local minimum
(b) The critical point with x∗ 6= 0 is (x∗ , y ∗ ) = (−a, −2).
At this point one has f ∗ (a) = f (−a, −2) = −a2 e−2 and df ∗ (a)/da = −2ae−2 .
On the other hand, for the function fˆ(x, y, a) = (x2 − axy)ey ,
∗
one has fˆ3′ (x, y, a) = −xyey and fˆ3′ (x∗ , y ∗ , a) = −x∗ y ∗ ey = −2ae−2 .
Thus the equation fˆ3′ (x∗ , y ∗ , a) = df ∗ (a)/da is true.
Of course, this is also what the envelope theorem tells us — see Eq. (13.7.2).
4. (a) Vt′ (t, x) = ft′ (t, x)e−rt − rf (t, x)e−rt = 0 and Vx′ (t, x) = fx′ (t, x)e−rt − 1 = 0, so at the
∗
optimum one has ft′ (t∗ , x∗ ) = rf (t∗ , x∗ ) and fx′ (t∗ , x∗ ) = ert .
(b) See the answer given in the book.
(c) V (t, x) = g(t)h(x)e−rt − x, so Vt′ = h(x)(g ′ (t) − rg(t))e−rt , Vx′ = g(t)h′ (x)e−rt − 1.
Moreover, Vtt′′ = h(x)(g ′′ (t) − 2rg ′ (t) + r2 g(t))e−rt , Vtx′′ = h′ (x)(g ′ (t) − rg(t))e−rt , and
′′ = g(t)h′′ (x)e−rt . Because the first-order condition g ′ (t∗ ) = rg(t∗ ) is satisfied at
Vxx
′′ < 0 provided that h′′ (x∗ ) < 0, and
(t∗ , x∗ ), there one has Vtx′′ = 0, as well as Vxx
∗
′′
∗
′′
∗
2
∗
−rt
Vtt = h(x )[g (t ) − r g(t )]e
< 0 provided that g ′′ (t∗ ) < r2 g(t∗ ). When both stated
′′ V ′′ − (V ′′ )2 > 0. These inequalities are sufficient
conditions are satisfied, one also has Vxx
tt
xt
to ensure that (t∗ , x∗ ) is a local maximum point.
√
√
√
∗
∗
(d) The first-order conditions in (b) reduce to e t /2 t∗ = re t , so t∗ = 1/4r2 , and also
1/(x∗ + 1) = e1/4r /e1/2r , or x∗ = e1/4r − 1. We check that the two conditions in (c) are
23
Theorem 13.2.1 cannot be used directly to prove optimality. However, it can be applied to the equivalent
problem of maximizing ln U . See Theorem 13.6.3.
55
satisfied. Obviously, h′′ (x∗ ) = −(1 + x∗ )−2 < 0. Moreover,
g ′′ (t∗ ) =
√
√
1
∗ √
√ e t ( t∗ − 1) = r2 (1 − 2r)e1/2r ,
∗
∗
4t t
whereas r2 g(t∗ ) = r2 e t = r2 e1/2r . Hence g ′′ (t∗ ) < r2 g(t∗ ) provided that r2 (1−2r) < r2 ,
which is true for all r > 0.
∗
6. (a) We need to have 1 + x2 y > 0. When x = 0, one has f (0, y) = 0. For x 6= 0, one has
1 + x2 y > 0 ⇐⇒ y > −1/x2 . The figure in the book shows part of the graph of f . Note
that f = 0 on the x-axis and on the y-axis.
(b) See the answer given in the book.
′′
(c) Here, f11
(x, y) =
2x
−x4
2y − 2x2 y 2 ′′
′′
,
f
(x,
y)
=
,
and
f
(x,
y)
=
.
22
(1 + x2 y)2 12
(1 + x2 y)2
(1 + x2 y)2
′′ (0, b) = 2b, f ′′ (0, b) = 0,
The second-order derivatives at all points of the form (0, b) are f11
12
′′ (0, b) = 0. Hence AC − B 2 = 0 at all the critical points, so the second-derivative
and f22
test tells us nothing about these points.
(d) See the answer given in the book.
13.4 Linear models with quadratic objectives
2. (a) See the answer given in the book.
(b) The new profit function is π̂ = −bp2 − dp2 + (a + βb)p + (c + βd)p − α − β(a + c).
The price that maximizes profits is easily seen to be p̂ = [a + c + β(b + d)]/2(b + d).
(c) In the case β = 0, the answers in part (a) simplify to p∗ = 21 a/b and q ∗ = 21 c/d,
with maximized profit π(p∗ , q ∗ ) = 41 a2 /b + 14 c2 /d − α. But when price discrimination is
prohibited, the answer in part (b) becomes p̂ = (a + c)/2(b + d), with maximized profit
(ad − bc)2
≥ 0.
π̂(p̂) = (a + c)2 /4(b + d)−α. The firm’s loss of profit is π(p∗ , q ∗ )− π̂(p̂) =
4bd(b + d)
Note that this loss is 0 if and only if ad = bc, in which case p∗ = q ∗ , so the firm wants to
charge the same price in each market anyway.
4. (a) The four data points are (x0 , y0 ) = (0, 11.29), (x1 , y1 ) = (1, 11.40), (x2 , y2 ) = (2, 11.49),
and (x3 , y3 ) = (3, 11.61), where x0 corresponds to 1970, etc.24 Using the method of least
squares set out in Example 4, we find that µx = 14 (0 + 1 + 2 + 3) = 1.5, µy = 41 (11.29 +
11.40+11.49+11.61) = 11.45, and σxx = 14 [(0−1.5)2 +(1−1.5)2 +(2−1.5)2 +(3−1.5)2 ] =
1.25. Moreover, σxy = 41 [(−1.5)(11.29 − 11.45) + (−0.5)(11.40 − 11.45) + (0.5)(11.49 −
11.45) + (1.5)(11.61 − 11.45)], which is equal to 0.13125, so formula (∗∗) implies that
â = σxy /σxx = 0.105 and b̂ = µy − âµx ≈ 11.45 − 0.105 · 1.5 = 11.29.
(b) With z0 = ln 274, z1 = ln 307, z2 = ln 436, and z3 = ln 524, the four data points are
(x0 , z0 ) = (0, 5.61), (x1 , z1 ) = (1, 5.73), (x2 , z2 ) = (2, 6.08), and (x3 , z3 ) = (3, 6.26). As
before, µx = 1.5 and σxx = 1.25. Moreover, µz = 14 (5.61 + 5.73 + 6.08 + 6.26) = 5.92 and
σxz ≈ 14 [(−1.5)(5.61−5.92)+(−0.5)(5.73−5.92)+(0.5)(6.08−5.92)+(1.5)(6.26−5.92)] =
0.2875. Hence ĉ = σxz /σxx = 0.23, dˆ = µz − ĉµx = 5.92 − 0.23 · 1.5 = 5.575.
24
The numbers yt are approximate, as are most subsequent results.
56
(c) If the time trends ln(GDP) = ax + b and ln(FA) = cx + d had continued, then FA would
have grown to equal 1% of GNP by the time x that solves ln(F A/GN P ) = ln 0.01 or
(c − a)x + d − b = ln 0.01. Hence x = (b − d + ln 0.01)/(c − a). Inserting the numerical
estimates found in parts (a) and (b) gives
x ≈ (11.29 − 5.575 − 4.605)/(0.23 − 0.105) = 1.11/0.125 = 8.88.
The goal would be reached in late 1978.
5. (a) The two firms’ combined profit is px + qy − (5 + x) − (3 + 2y), or substituting for x and
y, (p − 1)(29 − 5p + 4q) + (q − 2)(16 + 4p − 6q) − 8, which simplifies to 26p + 24q − 5p2 −
6q 2 + 8pq − 69. This is a concave function of p and q. The first-order conditions are the
two equations 26 − 10p + 8q = 0 and 24 − 12q + 8p = 0. The unique solution is p = 9,
q = 8, which gives a maximum. The corresponding production levels are x = 16 and
y = 4. Firm A’s profit is 123, whereas B’s is 21.
(b) Firm A’s profit is now πA (p) = (p − 1)(29 − 5p + 4q) − 5 = 34p − 5p2 + 4pq − 4q − 34, with
q fixed. This quadratic polynomial is maximized at p = pA (q) = 51 (2q + 17). Likewise,
firm B’s profit is now πB (q) = qy − 3 − 2y = 28q − 6q 2 + 4pq − 8p − 35, with p fixed. This
quadratic polynomial is maximized at q = qB (p) = 13 (p + 7).
(c) Equilibrium occurs where the price pair (p, q) satisfies the two equations p = pA (q) =
1
1
Substituting from the second equation
5 (2q+17) and q = qB (p) = 3 (p+7) simultaneously.
1
1
into the first yields p = 5 2 3 (p + 7) + 17 or, after clearing fractions, 15p = 2p + 14 + 51.
Hence prices are p = 5 and q = 4, whereas production levels are x = 20, y = 12, and
profits are 75 for A and 21 for B, respectively.
(d) Starting at (9, 8), first firm A moves to pA (8) = 33/5 = 6.6, then firm B answers by
moving to qA (6.6) = 13.6/3 ≈ 4.53, then firm A responds by moving to near pA (4.53) =
26.06/5 = 5.212, and so on. After the first horizontal move away from (9, 8), the process
keeps switching between moves vertically down from the curve p = pA (q), and moves
horizontally across to the curve q = qB (p), as shown in Fig. SM13.4.5. These moves
never cross either curve, and in the limit the process converges to the equilibrium (5, 4)
found in part (c).
q
10
y
p = pA (q)
4
8
(9, 8)
6
(3, 3)
q = qB (p)
4
(5, 4)
2
4
p
2
4
6
8
x
10 12 14
Figure SM13.5.2
Figure SM13.4.5
13.5 The Extreme Value Theorem
2. (a) The continuous function f is defined on a closed, bounded set S, as seen in Fig. SM13.5.2,
so the extreme value theorem ensures that f attains both a maximum and a minimum
57
over S.
Critical points are where: (i) f1′ (x, y) = 3x2 − 9y = 0, and (ii) f2′ (x, y) = 3y 2 − 9x = 0.
From (i), y = 31 x2 , which inserted into (ii) yields 31 x(x3 − 27) = 0. The only solutions are
x = 0 and x = 3. Thus the only critical point in the interior of S is (x, y) = (3, 3). We
proceed by examining the behaviour of f (x, y) along the boundary of S — i.e., along the
four edges of S.
(i) y = 0, x ∈ [0, 4]. Then f (x, 0) = x3 + 27, which has minimum at x = 0, and
maximum at x = 4.
(ii) x = 4, y ∈ [0, 4]. Then f (4, y) = y 3 − 36y + 91. The function g(y) = y 3 − 36y + 91,
√
y ∈ [0, 4] has g ′ (y) = 3y 2 − 36 = 0 at y = 12. Possible extreme points along (II)
√
are (4, 0), (4, 12), and (4, 4).
(iii) y = 4, x ∈ [0, 4]. Then f (x, 4) = x3 − 36x + 91, and as in (II) we see that possible
√
extreme points are (0, 4), ( 12, 4), and (4, 4).
(iv) x = 0, y ∈ [0, 4]. As in case (I) there are possible extreme points at (0, 0) and (0, 4).
This results in six candidates, where the function values are f (3, 3) = 0, f (0, 0) = 27,
√
√
√
f (4, 0) = f (0, 4) = 91, f (4, 12) = f ( 12, 4) = 91 − 24 12 ≈ 7.86, f (0, 0) = 27. The
conclusion follows.
(b) The constraint set S = (x, y) : x2 + y 2 ≤ 1 consists of points that lie on or inside a
circle around the origin with radius 1. This is a closed and bounded set, and f (x, y) =
x2 + 2y 2 − x is continuous. Therefore the extreme value theorem ensures that f attains
both a maximum and a minimum over S.
Critical points for f occur where fx′ (x, y) = 2x − 1 = 0 and fy′ (x, y) = 4y = 0. So the
only critical point for f is (x1 , y1 ) = (1/2, 0), which is an interior point of S.
An extreme point that does not lie in the interior of S must lie on the boundary of S,
that is, on the circle x2 + y 2 = 1. Along this circle we have y 2 = 1 − x2 , and therefore
f (x, y) = x2 + 2y 2 − x = x2 + 2(1 − x2 ) − x = 2 − x − x2
where x runs through the interval [−1, 1].25 The function g(x) = 2 − x − x2 has one
critical point in the interior of [−1, 1], namely x = −1/2, so any extreme values of g(x)
must occur either for this value of x or at one the endpoints ±1 of the interval [−1, 1].
Any extreme points for f (x, y) on the boundary of S must therefore be among the points
1
1 1√
1√
3 ), (x3 , y3 ) = (− , −
3 ), (x4 , y4 ) = (1, 0), (x5 , y5 ) = (−1, 0)
(x2 , y2 ) = (− ,
2 2
2
2
√
Now, f ( 21 , 0) = − 41 , f (− 12 , ± 12 3) = 49 , f (1, 0) = 0, and f (−1, 0) = 2. The conclusion in
the text follows.
3. The set S is shown in Fig. A13.5.3 in the book. It is clearly closed and bounded, so the
continuous function f has a maximum in S. The critical points are where ∂f /∂x = 9 −
12(x + y) = 0 and ∂f /∂y = 8 − 12(x + y) = 0. But 12(x + y) = 9 and 12(x + y) = 8 give
a contradiction. Hence, there are no critical points at all. The maximum value of f must
therefore occur on the boundary, which consists of five parts. Either the maximum value
25
It is a common error to overlook this restriction.
58
occurs at one of the five corners or “extreme points” of the boundary, or else at an interior
point of one of five straight “edges.” The function values at the five corners are f (0, 0) = 0,
f (5, 0) = −105, f (5, 3) = −315, f (4, 3) = −234, and f (0, 1) = 2.
We proceed to examine the behaviour of f at interior points along each of the five edges.
(i) y = 0, x ∈ (0, 5). The behaviour of f is determined by the function g1 (x) = f (x, 0) =
9x − 6x2 for x ∈ (0, 5). If this function of one variable has a maximum in (0, 5), it must
occur at a critical point where g1′ (x) = 9 − 12x = 0, and so at x = 3/4. We find that
g1 (3/4) = f (3/4, 0) = 27/8.
(ii) x = 5, y ∈ (0, 3). Define g2 (y) = f (5, y) = 45 + 8y − 6(5 + y)2 for y ∈ (0, 3). Here
g2′ (y) = −52 − 12y, which is negative throughout (0, 3), so there are no critical points
on this edge.
(iii) y = 3, x ∈ (4, 5). Define g3 (x) = f (x, 3) = 9x + 24 − 6(x + 3)2 for x ∈ (4, 5). Here
g3′ (x) = −27 − 12x, which is negative throughout (4, 5), so there are no critical points
on this edge either.
(iv) −x + 2y = 2, or y = x/2 + 1, with x ∈ (0, 4). Define the function g4 (x) = f (x, x/2 + 1) =
−27x2 /2 − 5x + 2 for x ∈ (0, 4). Here g4′ (x) = −27x − 5, which is negative in (0, 4), so
there are no critical points here.
(v) x = 0, y ∈ (0, 1). Define g5 (y) = f (0, y) = 8y − 6y 2 . Then g5′ (y) = 8 − 12y = 0 at
y = 2/3, with g5 (2/3) = f (0, 2/3) = 8/3.
After comparing the values of f at the five corners of the boundary and at the two points
found on the edges labelled (I) and (V) respectively, we conclude that the maximum value of
f is 27/8, which is achieved at (3/4, 0).
5. (a) First, f1′ (x, y) = e−x (1 − x)(y − 4)y = 0 when x = 1, or y = 0, or y = 4.
Second, f2′ (x, y) = 2xe−x (y − 2) = 0 when x = 0 or y = 2.
It follows that the critical points are (1, 2), (0, 0) and (0, 4).
′′ (x, y) = e−x (x − 2)(y 2 − 4y), f ′′ (x, y) = e−x (1 − x)(2y − 4), f ′′ = 2xe−x .
Moreover, f11
22
12
The critical points can be classified as follows:
(x, y)
(1, 2)
A
4e
−1
(0, 0)
0
(0, 4)
0
B
0
−4
4
C
2e
AC − B 2
−1
8e
0
−16
0
−2
−16
Type of point
Local minimum
Saddle point
Saddle point
(b) We show that the range of f is unbounded both above and below. Indeed, there is no
global maximum because f (1, y) = e−1 (y 2 − 4y) → ∞ as y → ∞. Nor is there any global
minimum because f (−1, y) = −e(y 2 − 4y) → −∞ as y → ∞.
(c) The set S is obviously bounded. The boundary of S consists of the four edges of the
rectangle, and all points on these line segments belong to S. Hence S is closed. Since f is
continuous, the extreme value theorem tells us that f has global maximum and minimum
points in S. These global extreme points must be either critical points if f in the interior
of S, or points on the boundary of S. The only critical point of f in the interior of S is
(1, 2). The function value at this point is f (1, 2) = −4e−1 ≈ −1.4715.
The four edges are most easily investigated separately:
59
(i) Along (I), y = 0 and f (x, y) = f (x, 0) is identically 0.
(ii) Along (II), x = 5 and f (x, y) = 5e−5 (y 2 − 4y) for y ∈ [0, 4]. This has its least value
for y = 2 and its greatest value for y = 0 and for y = 4. The function values are
f (5, 2) = −20e−5 ≈ −0.1348 and f (5, 0) = f (5, 4) = 0.
(iii) Along edge (III), y = 4 and f (x, y) = f (x, 4) = 0.
(iv) Finally, along (IV), x = 0 and f (x, y) = f (0, y) = 0.
Collecting all these results, we see that f attains its least value (on S) at the point (1, 2)
and its greatest value (namely 0) at all points of the line segments (I), (III) and (IV).
(d) y ′ = −
f1′ (x, y)
e−x (1 − x)(y − 4)y
(x − 1)(y − 4)y
=
−
=
= 0 when x = 1.
′
−x
f2 (x, y)
2xe (y − 2)
2x(y − 2)
13.6 The general case
Rz
2
et dt w.r.t. y, keeping z constant, can be
R
2
z 2
found using (9.3.7): it is −ey . The derivative of y et dt w.r.t. z, keeping y constant, can be
4. To calculate fx′ is routine. The derivative of
2
y
2
2
found using (9.3.6): it is ez . Thus fy′ = 2 − ey and fz′ = −3 + ez . Since each of the three
partials depends only on one variable and is 0 for two different values of that variable, there
are eight critical points (as indicated in the book).
13.7 Comparative statics and the Envelope Theorem
3. (a) With profits π as given in the problem, first-order conditions for a maximum are
2
′
= pK −1/3 − r = 0,
πK
3
1
πL′ = pL−1/2 − w = 0,
2
1
πT′ = pT −2/3 − q = 0.
3
Thus, K −1/3 = 3r/2p, L−1/2 = 2w/p, and T −2/3 = 3q/p. Raising each side of the equation K −1/3 = 3r/2p to the power of −3 yields K = (3r/2p)−3 = (2p/3r)3 = (8/27)p3 r−3 .
A similar method can be used to find L and T .
(b) Routine algebra: See the answer given in the book.
′ (K ∗ , L∗ ) = r using the product rule gives
5. (a) Differentiating pFK
′
′
FK
(K ∗ , L∗ )dp + p d[FK
(K ∗ , L∗ )] = dr.
′ (K ∗ , L∗ )] = F ′′ (K ∗ , L∗ ) dK ∗ + F ′′ (K ∗ , L∗ ) dL∗ .26
Moreover, d[FK
KK
KL
This explains the first displayed equation, replacing dK by dK ∗ and dL by dL∗ .
The second is derived in the same way.
(b) Rearrange the equation system by moving the differentials of the exogenous prices p, r,
and w to the right-hand side, while suppressing the notation indicating that the partials
are evaluated at (K ∗ , L∗ ):
′′
′′
′
pFKK
dK ∗ + pFKL
dL∗ = dr − FK
dp
′′
′′
dL∗ = dw − FL′ dp
dK ∗ + pFLL
pFLK
26
′
′
′
To see why, note that dg(K ∗ , L∗ ) = gK
(K ∗ , L∗ ) dK ∗ + gL
(K ∗ , L∗ ) dL∗ . Then let g = FK
.
60
′′ F ′′ − F ′′ F ′′ = F ′′ F ′′ − (F ′′ )2 . Using (2.4.2) and cancelling p,
Define ∆ = FKK
LL
KL LK
KK LL
KL
′′
′ F ′′ + F ′ F ′′
F ′′
−FKL
−FK
LL
L KL
dp + LL dr +
dw.
p∆
p∆
p∆
′′
′ F ′′
′′
−FL′ FKK
+ FK
−FLK
F ′′
LK
In the same way dL∗ =
dp +
dr + KK dw.
p∆
p∆
p∆
We can now read off the required partials.
we get dK ∗ =
(c) See the answer given in the book.27
6. (a)
(i) R1′ (x∗1 , x∗2 ) + σ = C1′ (x∗1 , x∗2 ): marginal revenue plus subsidy equals marginal cost.
(ii) R2′ (x∗1 , x∗2 ) = C2′ (x∗1 , x∗2 ) + τ = 0: marginal revenue equals marginal cost plus tax.
(b) See the answer given in the book, which introduces the notation
′′ 2
′′
′′
′′
′′
′′
) .
− C12
− C11
)(R22
− C22
) − (R12
D = (R11
(c) Taking the total differentials of (i) and (ii) yields
′′
′′
′′
′′
(R11
− C11
)dx∗1 + (R12
− C12
)dx∗2 = −dσ,
′′
′′
′′
′′
(R21
− C21
)dx∗1 + (R22
− C22
)dx∗2 = dτ
Solving for dx∗1 and dx∗2 yields, after rearranging,
dx∗1 =
′′ − C ′′ )dσ − (R′′ − C ′′ )dτ
−(R22
22
12
12
,
D
dx∗2 =
′′ − C ′′ )dσ + (R′′ − C ′′ )dτ
(R21
21
11
11
D
From the assumptions in the problem and the fact that D > 0 from (b), we find that the
partial derivatives are
′′ + C ′′
∂x∗1
−R22
22
=
> 0,
∂σ
D
′′ + C ′′
∂x∗1
−R12
12
=
> 0,
∂τ
D
′′
∂x∗2
R′′ − C21
= 21
< 0,
∂σ
D
′′
∂x∗2
R′′ − C11
= 11
< 0,
∂τ
D
Note that the signs of these partial derivatives accord with economic intuition. For
example, if the tax on good 2 increases, then the production of good 1 increases, while
the production of good 2 decreases.
′′ = R′′ and C ′′ = C ′′ .
(d) Follows from the expressions in (c) because R12
21
12
21
Review exercises for Chapter 13
2. (a) The profit function is π(Q1 , Q2 ) = 120Q1 + 90Q2 − 0.1Q21 − 0.1Q1 Q2 − 0.1Q22 .
First-order conditions for maximal profit are
π1′ (Q1 , Q2 ) = 120 − 0.2Q1 − 0.1Q2 = 0 and π2′ (Q1 , Q2 ) = 90 − 0.1Q1 − 0.2Q2 = 0.
We find (Q1 , Q2 ) = (500, 200).
′′ (Q , Q ) = −0.2 ≤ 0, π ′′ (Q , Q ) = −0.1, and π ′′ (Q , Q ) = −0.2 ≤ 0.
Moreover, π11
1
2
1
2
1
2
12
22
′′ π ′′ − (π ′′ )2 = 0.03 ≥ 0, (500, 200) maximizes profits.
Since also π11
22
12
27
′′
Recall that FLL
< 0 follows from (∗∗) in Example 13.3.3.
61
(b) The profit function is now π̂(Q1 , Q2 ) = P1 Q1 + 90Q2 − 0.1Q21 − 0.1Q1 Q2 − 0.1Q22 .
First-order conditions for maximal profit become
π̂1′ = P1 − 0.2Q1 − 0.1Q2 = 0 and π̂2′ = 90 − 0.1Q1 − 0.2Q2 = 0.
In order to induce the choice Q1 = 400,
the first-order conditions imply that P1 − 80 − 0.1Q2 = 0 and 90 − 40 − 0.2Q2 = 0.
It follows that Q2 = 250 and P1 = 105.
4. (a) Critical points are given by:
(i) f1′ (x, y) = 3x2 − 2xy = x(3x − 2y) = 0; and (ii) f2′ (x, y) = −x2 + 2y = 0.
From (i), x = 0 or 3x = 2y. If x = 0, then (ii) gives y = 0.
If 3x = 2y, then (ii) gives 3x = x2 , and so x = 0 or x = 3.
If x = 3, then (ii) gives y = x2 /2 = 9/2. So the critical points are (0, 0) and (3, 9/2).
(b) Critical points are given by: (i) f1′ (x, y) = ye4x
and (ii) f2′ (x, y) = xe4x
2 −5xy+y 2
2 −5xy+y 2
(2y 2 − 5xy + 1) = 0.
(8x2 − 5xy + 1) = 0;
If y = 0, then (i) is satisfied and (ii) holds only when x = 0.
If x = 0, then (ii) is satisfied and (i) holds only if y = 0.
Thus, in addition to (0, 0), any other critical point must satisfy both 8x2 − 5xy + 1 = 0
and 2y 2 − 5xy + 1 = 0.
Subtracting the second of these equations from the first yields 8x2 = 2y 2 , or y = ±2x.
Inserting y = −2x into 8x2 − 5xy + 1 = 0 yields 18x2 + 1 = 0, which has no solutions.
√
Inserting y = 2x into 8x2 − 5xy + 1 = 0 yields −2x2 + 1 = 0, so x = ± 12 2.
√ √
√
√
We conclude that the critical points are: (0, 0) and ( 21 2, 2 ), (− 21 2, − 2 ).
(c) Critical points occur where: (i) f1′ (x, y) = 24xy − 48x = 24x(y − 2) = 0;
and (ii) f2′ (x, y) = 12y 2 + 12x2 − 48y = 12(x2 + y 2 − 4y) = 0.
From (i) x = 0 or y = 2.
If x = 0, then (ii) gives y(y − 4) = 0, so y = 0 or y = 4. So (0, 0) and (0, 4) are critical
points.
If y = 2, then (ii) gives x2 − 4 = 0, so x = ±2. Hence (2, 2) and (−2, 2) are also critical
points.
6. (a) With π = p(K a + Lb + T c ) − rK − wL − qT , the first-order conditions for (K ∗ , L∗ , T ∗ ) to
maximize π are
′
= pa(K ∗ )a−1 − r = 0,
πK
πL′ = pb(L∗ )a−1 − w = 0,
πT′ = pc(T ∗ )a−1 − q = 0
Hence, K ∗ = (ap/r)1/(1−a) , L∗ = (bp/w)1/(1−a) , T ∗ = (cp/q)1/(1−a) .
(b) π ∗ = Γ + terms that do not depend on r, where Γ = p(ap/r)a/(1−a) − r(ap/r)1/(1−a) .
Some algebraic manipulations yield
a a/(1−a)
Γ=
p1/(1−a) − (ap)1/(1−a) r−a/(1−a)
r
= aa/(1−a) − a1/(1−a) p1/(1−a) r−a/(1−a)
= (1 − a)aa/(1−a) p1/(1−a) r−a/(1−a) .
Then,
ap 1/(1−a)
−1
∂π ∗
∂Γ
=
= −aaa/(1−a) p1/(1−a) r−a/(1−a) = −a1/(1−a) p1/(1−a) r 1−a = −
.
∂r
∂r
r
62
(c) We apply (13.7.2) to this case, where π(K, L, T, p, r, w, q) = pQ − rK − wL − qT with
Q = K a + Lb + T c , and π ∗ (p, r, w, q) = pQ∗ − rK ∗ − wL∗ − qT ∗ . With the partial
derivatives of π evaluated at (K ∗ , L∗ , T ∗ , p, r, w, q) where output is Q∗ , one should have
′ = −L∗ , and ∂π ∗ /∂w = π ′ =
∂π ∗ /∂p = πp′ = Q∗ , ∂π ∗ /∂r = πr′ = −K ∗ , ∂π ∗ /∂w = πw
w
−T ∗ . From (b), we have the second property. The other three equations can be verified
by rather tedious algebra in a similar way.
′′ (x, y) = 2 − 6x, f ′′ (x, y) = −1, f ′′ (x, y) =
8. (a) f1′ (x, y) = 2x − y − 3x2 , f2′ (x, y) = −2y − x, f11
12
22
−2. Critical points occur where 2x − y − 3x2 = 0 and −2y − x = 0. The last equation
yields y = −x/2, which inserted into the first equation gives 25 x − 3x2 = 0. It follows that
there are two critical points, (x1 , y1 ) = (0, 0) and (x2 , y2 ) = (5/6, −5/12). These points
are classified in the following table:
(x, y)
A
B
C
(0, 0)
2
5
)
( 56 , − 12
−3
−1
−2
−1
−2
AC − B 2
Type of point
5
Local maximum
−5
Saddle point
′′ ≤ 0, f ′′ ≤ 0, and f ′′ f ′′ − (f ′′ )2 ≥ 0 — i.e., where 2 − 6x ≤ 0,
(b) f is concave where f11
22
11 22
12
−2 ≤ 0, and (2 − 6x)(−2) − (−1)2 ≥ 0. These conditions are equivalent to x ≥ 1/3 and
x ≥ 5/12. Since 5/12 > 1/3, the function f is concave in the set S consisting of all (x, y)
where x ≥ 5/12.
(c) The critical point (x2 , y2 ) = (5/6, −5/12) found in (a) belongs to S. Since f is concave
25
25
125
125
in S, this is a (global) maximum point for f in S, and fmax = 25
36 − 144 + 72 − 216 = 432 .
9. (a) Critical points are where x − 1 = −ay and a(x − 1) = y 2 − 2a2 y. Multiplying the
first equation by a gives −a2 y = a(x − 1) = y 2 − 2a2 y, by the second equation. Hence
a2 y = y 2 , implying that y = 0 or y = a2 . Since x = 1 − ay, the critical points are (1, 0)
and (1 − a3 , a2 ).28
(b) The function value at the critical point in (a) is
1
2 (1
− a3 )2 − (1 − a3 ) + a3 (−a3 ) − 31 a6 + a2 · a4 = − 12 + 61 a6 ,
whose derivative w.r.t. a is a5 . On the other hand, the partial derivative of f w.r.t. a,
keeping x and y constant, is ∂f /∂a = y(x − 1) + 2ay 2 . Evaluated at x = 1 − a3 , y = a2 ,
this partial derivative is also a5 , thus confirming the envelope theorem.
′′ = 1, f ′′ = −2y + 2a2 , f ′′ = a, and f ′′ f ′′ − (f ′′ )2 = a2 − 2y. Thus f is convex if and
(c) f11
22
12
11 22
12
only if −2y + 2a2 ≥ 0 and −2y + a2 ≥ 0, which is equivalent to a2 ≥ y and a2 ≥ 2y. It
follows that f (x, y) is convex in that part of the xy-plane where y ≤ 12 a2 .
14 Constrained Optimization
14.1 The Lagrange Multiplier method
4. (a) With L(x, y) = x2 + y 2 − λ(x + 2y − 4), the first-order conditions are L′1 = 2x − λ = 0
and L′2 = 2y − 2λ = 0. From these equations we get 2x = y, which inserted into the
constraint gives x + 4x = 4. So x = 4/5 and y = 2x = 8/5, with λ = 2x = 8/5.
28
Since you were asked only to show that (1 − a3 , a2 ) is a critical point, it would suffice to verify that it makes
both partials equal to 0.
63
(b) The same method as in (a) gives 2x − λ = 0 and 4y − λ = 0, so x = 2y. From the
constraint we get x = 8 and y = 4, with λ = 16.
(c) The first-order conditions imply that 2x + 3y = λ = 3x + 2y, which gives x = y. So the
solution is (x, y) = (50, 50) with λ = 250.
9. (a) With L = xa + y − λ(px + y − m), the first-order conditions for (x∗ , y ∗ ) to solve the
problem are
L′x = a(x∗ )a−1 − λp = 0,
L′y = 1 − λ = 0.
Thus λ = 1, and x∗ = x∗ (p, m) = kp−1/(1−a) where k = a1/(1−a) .
Then y ∗ = y ∗ (p, m) = m − kp−a/(1−a) .
(b) ∂x∗ /∂p = −x∗ /p(1 − a) < 0, ∂x∗ /∂m = 0, ∂y ∗ /∂p = ax∗ /(1 − a) > 0, and ∂y ∗ /∂m = 1.
(c) The optimal expenditure on good x is px∗ (p, m) = kp−a/(1−a) , so Elp px∗ (p, m) = −a/(1−
a) < 0. In particular, the expenditure on good x will decrease as its price increases.
√
(d) We see that x∗ = (1/2p)2 , y ∗ = m − 1/4p, so U ∗ (p, m) = x∗ + y ∗ = (1/2p) + m − 1/4p =
m + 1/4p, and the required identity is obvious.
10. (a) With L(x, y) = 100 − e−x − e−y − λ(px + qy − m), the first-order conditions L′x =
L′y = 0 imply that e−x = λp and e−y = λq. Hence, x = − ln(λp) = − ln λ − ln p
and y = − ln λ − ln q. Inserting these expressions for x and y into the constraint yields
−p(ln λ + ln p) − q(ln λ + ln q) = m and so ln λ = −(m + p ln p + q ln q)/(p + q). Therefore
x(p, q, m) =
m + q ln(q/p)
p+q
and
y(p, q, m) =
m + p ln(p/q)
.
p+q
(b) x(tp, tq, tm) = [tm + tq ln(tq/tp)]/(tp + tq) = x(p, q, m), so x is homogeneous of degree 0.
In the same way we see that y(p, q, m) is homogeneous of degree 0.
14.2 Interpreting the Lagrange Multiplier
√
4. (a) With L(x, y) = x + y − λ(x + 4y − 100), the first-order conditions for (x∗ , y ∗ ) to solve
the problem are:
√
(i) ∂L/∂x = 1/2 x∗ − λ = 0,
(ii) ∂L/∂y = 1 − 4λ = 0.
√
From (ii), λ = 1/4, which inserted into (i) yields x∗ = 2, so x∗ = 4. Then y ∗ = 25− 41 4 =
√
24, and maximal utility is U ∗ = x∗ + y ∗ = 26.
(b) Denote the new optimal values of x and y by x̂ and ŷ. If 100 is changed to 101, still
λ = 1/4
√ and x̂ = 4. The constraint now gives 4+4ŷ = 101, so that ŷ = 97/4 = 24.25, with
Û = x̂ + ŷ = 26.25. The increase in maximum utility is therefore Û − U ∗ = 0.25 = λ.29
√
(c) The necessary conditions for optimality are now ∂L/∂x = 1/2 x∗ − λp = 0, ∂L/∂y =
√
1 − λq = 0. Proceeding in the same way as in (a), we find λ = 1/q, x∗ = q/2p, and so
x∗ = q 2 /4p2 , with y ∗ = m/q − q/4p.30
29
In general, the increase in utility is approximately equal to the value of the Lagrange multiplier.
Note that y ∗ > 0 ⇐⇒ m > q 2 /4p. Also, note that if we solve the constraint for y, the utility function is
√
√
u(x) = x + (m − px)/q. We see that u′ (x) = 1/2 x − p/q = 0 for x∗ = q 2 /4p2 and u′′ (x) = −(1/4)x−3/2 < 0 when
x > 0. So we have found the maximum.
30
64
5. (a) The first-order conditions given in the main text imply that px∗ = pa + α/λ and qy ∗ =
qb + β/λ. Substituting these into the budget constraint gives m = px∗ + qy ∗ = pa + qb +
(α + β)/λ = pa + qb + 1/λ, so 1/λ = m − (pa + qb). The expressions given in (∗∗) are
now easily established.31
(b) With U ∗ as given in the answer provided in the text, differentiating partially while remembering that α + β = 1 gives
∂U ∗
α
β
1
=
+
=
= λ > 0.
∂m
m − (pa + qb) m − (pa + qb)
m − (pa + qb)
−αa
α
−βa
−a
α
α
∂U ∗
=
− +
=
− = −aλ − ,
∂p
m − (pa
m − (pa + qb) p
p
+ qb) p m − (pa + qb)
α
∂U ∗ ∗
α
x = −λ a +
whereas −
= −aλ − , so ∂U ∗ /∂p = −∂U ∗ /∂m · x∗ .
∂m
λp
p
The last equality is shown in the same way.
Moreover,
6. A formula for f (x, T ) is
x
Z
0
T
−t3 + (αT 2 + T − 1)t2 + (T − αT 3 )t dt = x
T
0
− 14 t4 + (αT 2 + T − 1) 31 t3 + (T − αT 3 ) 21 t2
1
1
1
1
= − αxT 5 + xT 4 + xT 3 = xT 3 (2 + T − 2αT 2 )
6
12
6
12
RT
A similar, but easier, calculation shows that g(x, T ) = 0 (xtT − xt2 ) dt = 61 xT 3 . The
Lagrangian for the producer’s problem is
1
1 3
L = xT 3 (2 + T − 2αT 2 ) − λ
xT − M .
12
6
The two first-order conditions are
1 3
1
T (2 + T − 2αT 2 ) − λT 3 = 0
12
6
and
1
1
xT 2 (6 + 4T − 10αT 2 ) − λxT 2 = 0.
12
2
These equations imply that λ = 12 (2 + T − 2αT 2 ) = 61 (6 + 4T − 10αT 2 ). It follows that
4αT 2 = T . One solution is T = 0, but this is inconsistent with the constraint 61 xT 3 = M .
Hence, the solution we are interested is T = 1/4α, implying that λ = 1 + 1/16α. Substituting
into the constraint g(x, T ) = M determines x = 6M T −3 = 384M α3 . Because xT 3 = 6M ,
the maximum profit is f ∗ (M ) = M + M/8α − αM/16α2 = M + M/16α, whose derivative
w.r.t. M is indeed λ.32
14.3 Multiple solution candidates
1. (a) With L(x, y) = 3xy − λ(x2 + y 2 − 8), the first-order conditions are L′1 = 3y − 2λx = 0
and L′2 = 3x − 2λy = 0. These can be rewritten as: (i) 3y = 2λx; (ii) 3x = 2λy.
31
We can interpret a and b as minimum subsistence quantities of the two goods, in which case the assumption
pa + qb < m means that the consumer can afford to buy (a, b).
32
We note that the maximum by itself is much easier to find if one substitutes the constraint M = 61 xT 3 into
the objective function f (x, T ), which then becomes the function −αM T 2 + 21 M T + M of T alone, with α and M
as parameters. The first-order condition for T to maximize this expression is −2αM T + 21 M = 0, implying that
T = 1/4α. However, this method does not find λ.
65
If x = 0, then (i) gives y = 0; conversely, if y = 0, then (ii) gives x = 0. But (x, y) = (0, 0)
does not satisfy the constraint. Hence x 6= 0 and y 6= 0. Equating the ratio of the lefthand sides of (i) and (ii) to the ratio of their right-hand sides, we get y/x = x/y or
x2 = y 2 . Finally, using the constraint gives x2 = y 2 = 4.
The four solution candidates are therefore (2, 2) and (−2, −2) with λ = 3/2, as well as
(2, −2) and (−2, 2) with λ = −3/2. The corresponding function values are f (2, 2) =
f (−2, −2) = 12 and f (2, −2) = f (−2, 2) = −12.
Because f is continuous and the constraint set is a circle, which is closed and bounded, the
extreme value theorem implies that a maximum and a minimum do exist. The function
values tell us that (2, 2) and (−2, −2) solve the maximization problem, whereas (−2, 2)
and (2, −2) solve the minimization problem.
(b) With L = x + y − λ(x2 + 3xy + 3y 2 − 3), the first-order conditions are
1 − 2λx − 3λy = 0
and
1 − 3λx − 6λy = 0.
These equations give us 1 = 2λx + 3λy = 3λx + 6λy. In particular, λ(3y + x) = 0. Here
λ = 0 is impossible, so x = −3y. Inserting this into the constraint reduces it to 3y 2 = 3,
with solutions y = ±1. So the first-order conditions give two solution candidates:
(x, y, λ) = (3, −1, 31 )
and
(x, y, λ) = (−3, 1, − 31 ).
The objective function is continuous and the constraint curve is closed and bounded —
actually, an ellipse — see (5.5.5). So the extreme value theorem ensures that both a
maximum and minimum exist. The two function values f (3, −1) = 2 and f (−3, 1) = −2
tell us that the maximum is at (3, −1), the minimum at (−3, 1).
2. (a) With L = x2 + y 2 − 2x + 1 − λ(x2 + 4y 2 − 16), the first-order conditions are
(i) 2x − 2 − 2λx = 0;
(ii) 2y − 8λy = 0.
Equation (i) implies that x 6= 0 and then λ = 1 − 1/x, whereas equation (ii) shows that
y = 0 or λ = 1/4. If y = 0, then x2 = 16 − 4y 2 = 16, so x = ±4, which then gives
λ = 1 ∓ 1/4. If y 6= 0, then λ = 1/4 and (i) gives 2x − 2 − x/2 = 0,pso x = 4/3. The
√
constraint x2 +4y 2 = 16 now yields 4y 2 = 16−16/9 = 128/9, so y = ± 32/9 = ±4 2/3.
Thus, there are four solution candidates:
√
(a) (x, y, λ) = (4, 0, 3/4);
(b) (x, y, λ) = (4/3, 4 2/3, 1/4);
√
(c) (x, y, λ) = (−4, 0, 5/4);
(d) (x, y, λ) = (4/3, −4 2/3, 1/4).
Of these four, checking function values shows that (a) and (b) both give a maximum,
whereas (c) and (d) both give a minimum.
(b) The Lagrangian is L = ln (2 + x2 ) + y 2 − λ(x2 + 2y − 2). Hence, the necessary first-order
conditions for (x, y) to be a minimum point are
∂
∂
L = 2x/(2 + x2 ) − 2λx = 0; (ii)
L = 2y − 2λ = 0; (iii) x2 + 2y = 2.
∂x
∂y
From (i) we get x 1/(2 + x2 ) − λ = 0, so x = 0 or λ = 1/(2 + x2 ).
(i)
66
Now if x = 0, then (iii) gives y = 1, so (x1 , y1 ) = (0, 1) is a first solution candidate.
Alternatively, if x 6= 0, then (ii) and then (i) imply that y = λ = 1/(2 + x2 ). In this case,
inserting y = 1/(2 + x2 ) into (iii) gives
√
4
x2 + 2/(2 + x2 ) = 2 ⇐⇒ 2x2 + x4 + 2 = 4 + 2x2 ⇐⇒ x4 = 2 ⇐⇒ x = ± 2.
√
From (iii), y = 1 − 12 x2 = 1 − 12 2. This gives us the two other solution candidates
√
√
√
√
4
4
(x3 , y3 ) = (− 2, 1 − 21 2).
(x2 , y2 ) = ( 2, 1 − 21 2);
Comparing function values, we see that
f (x1 , y1 ) = f (0, 1) = ln 2 + 1 ≈ 1.69
√
√
√
f (x2 , y2 ) = f (x3 , y3 ) = ln (2 + 2) + (1 − 12 2)2 = ln (2 + 2) +
3
2
−
√
2 ≈ 1.31.
Hence, the minimum points for f (x, y) subject to the constraint are (x2 , y2 ) and (x3 , y3 ).
4. (a) With L = 24x − x2 + 16y − 2y 2 − λ(x2 + 2y 2 − 44), the two first-order conditions are:
(i) L′1 = 24 − 2x − 2λx = 0;
(ii) L′2 = 16 − 4y − 4λy = 0.
From (i) x(1 + λ) = 12 and from (ii) y(1 + λ) = 4. Eliminating λ from (i) and (ii) we
get x = 3y = 12/(1 + λ), with λ 6= −1. Inserted into the constraint, 11y 2 = 44, so
y = ±2, and then x = ±6. So there are two solution candidates, (x, y) = (6, 2) and
(−6, −2), both with λ = 1. Computing the objective function at these two points, the
only possible maximum is at (x, y) = (6, 2). Since the objective function is continuous
and the constraint curve is closed and bounded (an ellipse), the extreme value theorem
assures us that there is indeed a maximum at this point.
(b) According to (14.2.3) the change is approximately λ · 1 = 1.
14.4 Why the Lagrange method works
4. Before trying to find the minimum, consider the graph of the curve g(x, y) = 0, as shown in
Fig. SM14.4.4. It consists of three pieces:
√
(i) the continuous curve y = x(x + 1) in the positive quadrant;
√
(ii) the continuous curve y = − x(x + 1), which is the reflection of curve (i) about the
x-axis;
(iii) the isolated point (−1, 0). The problem is to minimize the square of the distance d from
the point (−2, 0) to a point on this graph. The minimum of f obviously occurs at the
isolated point (−1, 0), with f (−1, 0) = 1.
With the Lagrangian L = (x + 2)2 + y 2 − λ[y 2 − x(x + 1)2 ], we have
L′1 = 2(x + 2) + λ[(x + 1)2 + 2x(x + 1)]
and L′2 = 2y(1 − λ). Note that L′2 = 2y(1 − λ) = 0 only if λ = 1 or y = 0. For λ = 1, we find
that L′1 = 3(x + 1)2 + 2 > 0 for all x. For y = 0, the constraint gives x = 0 or x = −1. At
x = 0, we have L′1 = 4 + λ = 0 whereas L′1 = 2 at x = −1.
67
y
6
5
4
3
2
d 1
−3−2−1
−1
−2
−3
−4
−5
−6
y 2 = x(x + 1)2
(x, y)
1 2 3 4
x
Figure SM14.4.4
Thus the Lagrange multiplier method produces a unique solution candidate (x, y) = (0, 0)
with λ = −4, which does correspond to a local minimum. The global minimum at (−1, 0),
however, fails to satisfy the first-order conditions L′1 = L′2 = 0 for any value of λ, because
L′1 = 2 at this point. So the Lagrange multiplier method cannot locate this minimum. Note
that at (−1, 0) both g1′ (−1, 0) and g2′ (−1, 0) are 0.
14.5 Sufficient conditions
4. With L = xa + y a − λ(px + qy − m), the first-order conditions are:
L′1 = axa−1 − λp = 0
and
L′2 = ay a−1 − λq = 0.
If λ = 0, then (x, y) = (0, 0), which does not satisfy the constraint. It follows that λ 6= 0
and then x = (λp/a)1/(a−1) , y = (λq/a)1/(a−1) . Inserting these values of x and y into the
budget constraint gives (λ/a)1/(a−1) (pa/(a−1) + q a/(a−1) ) = m. To reduce notation, define
R = pa/(a−1) + q a/(a−1) as in the answer in the book. Then we have (λ/a)1/(1−a) = m/R.
Therefore x = mp1/(a−1) /R and y = mq 1/(a−1) /R.
14.6 Additional variables and constraints
7. The Lagrangian is L = x + y − λ(x2 + 2y 2 + z 2 − 1) − µ(x + y + z − 1), which has critical
points when
(i) L′x = 1 − 2λx − µ = 0;
(ii) L′y = 1 − 4λy − µ = 0;
(iii) L′z = −2λz − µ = 0.
From (ii) and (iii) we get 1 = λ(4y − 2z), and in particular λ 6= 0. From (i) and (ii),
λ(x−2y) = 0 and so x = 2y. Substituting this value for x into the constraints gives 6y 2 +z 2 = 1
and 3y + z = 1. Thus z = 1 − 3y, implying that 1 = 6y 2 + (1 − 3y)2 = 15y 2 − 6y + 1. Hence
y = 0 or y = 2/5, implying that x = 0 or 4/5, and that z = 1 or −1/5. The only two solution
candidates are (x, y, z) = (0, 0, 1) with λ = −1/2, µ = 1, and (x, y, z) = (4/5, 2/5, −1/5)
68
with λ = 1/2, µ = 1/5. Because x + y is 0 at (0, 0, 1) and 6/5 at (4/5, 2/5, −1/5), these are
respectively the minimum and the maximum.33
8. (a) For the given Cobb–Douglas utility function, one has Uj′ (x) = αj U (x)/xj . So (14.6.6)
with k = 1 implies that pj /p1 = Uj′ (x)/U1′ (x) = αj x1 /α1 xj . Thus pj xj = (aj /a1 )p1 x1 .
Inserting this into the budget constraint for j = 2, . . . , n, gives p1 x1 + (a2 /a1 )p1 x1 +
· · · + (an /a1 )p1 x1 = m, which implies that p1 x1 = a1 m/(a1 + · · · + an ). Similarly,
pj xj = aj m/(a1 + · · · + an ) for k = 1, . . . , n.
(b) From (14.6.6) with k = 1, we get xja−1 /x1a−1 = pj /p1 and so xj /x1 = (pj /p1 )−1/(1−a) , or
pj xj /p1 x1 = (pj /p1 )1−1/(1−a) = (pj /p1 )−a/(1−a) . Inserting this into the budget constraint
for j = 2, . . . , n, gives
"
−a/(1−a)
−a/(1−a) #
p2
pn
p1 x 1 1 +
=m
+ ··· +
p1
p1
−a/(1−a)
mp
so p1 x1 = P 1 −a/(1−a) . Arguing similarly for each j, we get
n
i=1 pi
−a/(1−a)
mpj
pj x j = P
n
−a/(1−a)
i=1 pi
14.7 Comparative statics
for j = 1, . . . , n.
2. Here, L = x + 4y + 3z − λ(x2 + 2y 2 + 13 z 2 − b), so necessary first-order conditions are:
(i) L′1 = 1 − 2λx = 0
(ii) L′2 = 4 − 4λy = 0
2
(iii) L′3 = 3 − λz = 0
3
It follows that λ 6= 0, and so x = 1/2λ, y = 1/λ, z = 9/2λ. Inserting these values into √
the
constraint yields [(1/4) + 2 + (27/4)]λ−2 = b and so λ2 = 9/b, implying√that λ = ±3/ b.
The value of the objective function is x + 4y + 3z = 18/λ, √so λ = −3/ b determines the
minimum point. This is (x, y, z) = (a, 2a, 9a), where a = − b/6. For the last verification,
see the answer in the book.
4. (a) With L = x2 + y 2 + z − λ(x2 + 2y 2 + 4z 2 − 1), necessary conditions are:
(i) ∂L/∂x = 2x − 2λx = 0
(ii) ∂L/∂y = 2y − 4λy = 0
(iii) ∂L/∂z = 1 − 8λz = 0
From (i), 2x(1 − λ) = 0, so there are two possibilities: x = 0 or λ = 1.
(A) Suppose x = 0. From (ii), 2y(1 − 2λ) = 0, so y = 0 or λ = 1/2.
(A.1) If y = 0, then the constraint gives 4z 2 = 1, so z 2 = 1/4, or z = ±1/2. Equation
(iii) gives λ = 1/8z, so we have two solution candidates: P1 = (0, 0, 1/2) with
λ = 1/4; and P2 = (0, 0, −1/2) with λ = −1/4.
33
The two constraints determine the curve in three dimensions which is the intersection of an ellipsoid (see Fig.
11.4.2) and a plane. Because an ellipsoid is a closed bounded set, so is this curve. By the extreme value theorem,
the continuous function x + y does attain a maximum and a minimum over this closed bounded set.
69
(A.2) If λ = 1/2, then (iii) gives z = 1/8λ = 1/4. It follows frompthe constraint that
√
2y 2 = 3/4 (recall that we assumed x = 0), and hence y = ± 3/8 = ± 6/4. So
√
√
new candidates are: P3 = (0, 6/4, 1/4) with λ = 1/2; and P4 = (0, − 6/4, 1/4)
with λ = 1/2.
(B) Suppose λ = 1. Equation (iii) yields z = 1/8, and (ii) gives y = 0. From the
√
constraint, x2 = 15/16, so x = ± 15/4. This gives us two more solution andidates:
√
√
P5 = ( 15/4, 0, 1/8) with λ = 1; and P6 = (− 15/4, 0, 1/8) with λ = 1.
For k = 1, 2, . . . , 6, let fk denote the value of the criterion function f at the candidate
point Pk . Routine calculation shows that f1 = 1/2, f2 = −1/2, f3 = f4 = 5/8, and
f5 = f6 = 17/16. It follows that both P5 and P6 solve the maximization problem,
whereas P2 solves the minimization problem.
(b) See the answer given in the book.
5. The Lagrangian is L = rK + wL − λ(K 1/2 L1/4 − Q), so first-order necessary conditions for
(K ∗ , L∗ ) to solve the problem are:
(i) L′K = r − 12 λ(K ∗ )−1/2 (L∗ )1/4 = 0
and
(ii) L′L = w − 41 λ(K ∗ )1/2 (L∗ )−3/4 = 0
together with the constraint: (iii) (K ∗ )1/2 (L∗ )1/4 = Q.
Together (i) and (ii) imply that r/w = 2L∗ /K ∗ and so L∗ = rK ∗ /2w. Inserting the latter
into (iii) gives Q = (K ∗ )1/2 (rK ∗ /2w)1/4 = (K ∗ )3/4 2−1/4 r1/4 w−1/4 . Solving for K ∗ gives
the answer in the text. The answers for L∗ and C ∗ = rK ∗ + wL∗ follow if we observe that
21/3 = 2 · 2−2/3 . Verifying (∗) in Example 14.7.3 is easy.
14.8 Nonlinear programming: a simple case
2
5. (a) With the Lagrangian L(x, y) = 2 − (x − 1)2 − ey − λ(x2 + y 2 − a), the Kuhn–Tucker
conditions are:
(i) −2(x−1)−2λx = 0;
2
(ii) −2yey −2λy = 0;
(iii) λ ≥ 0, with λ = 0 if x2 + y 2 < a.
2
From (i), x = (1 + λ)−1 . Moreover, (ii) reduces to y(ey + λ) = 0, and so y = 0, because
2
ey + λ is always positive.
(I) Assume that λ = 0. Then equation (i) gives x = 1. In this case we must have
a ≥ x2 + y 2 = 1.
√
(II) Assume that λ > 0. Then (iii) gives x2 + y 2 = a, and so x = ± a (remember that
√
y = 0). Because x = 1/(1 + λ) and λ > 0 we must have 0 < x < 1, so x = a
and a = x2 < 1. It remains to find the value of λ and check that it is > 0. From
√
equation (i) we get λ = 1/x − 1 = 1/ a − 1 > 0.
Conclusion: The only point that satisfies the Kuhn–Tucker conditions is (x, y) = (1, 0)
√
√
if a ≥ 1 and ( a, 0) if 0 < a < 1. The corresponding value of λ is 0 or 1/ a − 1,
respectively.
2
(b) Since L′11 = −2 − 2λ < 0, L′22 = −ey (2 + 4y 2 ) − 2λ < 0, and L′12 = 0, the Lagrangian
L(x, y) is concave, so we have found the solution in both cases(I) and (ii) in part (a)..
70
√
√
√
(c) If a ∈ (0, 1) we have f ∗ (a) = f ( a, 0) = 2 − ( a − 1)2 − 1 = 2 a − a, and for a ≥ 1 we
get f ∗ (a) = f (1, 0) = 1. The derivative of f ∗ is as given in the book, but note that in
order to find the derivative df ∗ (a)/da when a = 1, we need to show that the right and
left derivatives (see Section 7.9 in the book)
(f ∗ )′ (1+ ) = lim
h→0+
f ∗ (1 + h) − f ∗ (1)
h
and
(f ∗ )′ (1− ) = lim
h→0−
f ∗ (1 + h) − f ∗ (1)
h
exist and are equal. But the left and right derivatives are respectively equal to the
√
derivatives of the differentiable functions g− (a) = 2 a − a and g+ (a) = 1 at a = 1, which
are both 0. Hence (f ∗ )′ (1) exists and equals 0.
14.9 Multiple inequality constraints
2. The Lagrangian is L = α ln x + (1 − α) ln y − λ(px + qy − m) − µ(x − x̄), and the Kuhn–Tucker
conditions for (x∗ , y ∗ ) to solve the problem are:
α
(iii) λ ≥ 0, with λ = 0 if px∗ + qy ∗ < m
(i) L′1 = ∗ − λp − µ = 0
x
1−α
− λq = 0
(iv) µ ≥ 0, with µ = 0 if x∗ < x̄.
(ii) L′2 =
y∗
Since α ∈ (0, 1), (ii) implies that λ > 0 and so (iii) entails px∗ + qy ∗ = m.
Suppose µ = 0. Then (i) and (ii) imply that α/px∗ = (1 − α)/qy ∗ , so qy ∗ = (1 − α)px∗ /α.
Then the budget constraint implies that px∗ + (1 − α)px∗ /α = m, so x∗ = mα/p and then
y ∗ = (1 − α)m/q, with λ = 1/m. This solution is valid as long as x∗ ≤ x̄, that is m ≤ px̄/α.
Suppose µ > 0. Then x∗ = x̄ and y ∗ = m/q − px̄/q = (m − px̄)/q, with λ = (1 − α)/(m − px̄)
and µ = α/x̄ − λp = (αm − px̄)/x̄(m − px̄). Note that if m > px̄/α, then m > px̄ since α < 1.
We conclude that if m > px̄/α, then λ and µ are both positive and so (i)–(iv) are all satisfied.
Since L′11 = −α/x2 < 0, L′22 = −α/y 2 < 0, and L′12 = 0, the Lagrangian L(x, y) is concave,
so we have found the solution in both cases.
3. (a) See the answer given in the book.
(b) With the constraints g1 (x, y) = −x−y ≤ −4, g2 (x, y) = −x ≤ 1, and g3 (x, y) = −y ≤ −1,
the Lagrangian is L = x + y − ex − ex+y − λ1 (−x − y + 4) − λ2 (−x − 1) − λ3 (−y + 1).
Formulating the complementary slackness conditions as in (14.8.5), the Kuhn–Tucker
conditions are that there exist nonnegative numbers λ1 , λ2 , and λ3 such that:
(i) L′x = 1 − ex − ex+y + λ1 + λ2 = 0
(ii) L′y = 1 − ex+y + λ1 + λ3 = 0
(iii) λ1 (−x − y + 4) = 0
(iv) λ2 (−x − 1) = 0
(v) λ3 (−y + 1) = 0
From (ii), ex+y = 1+λ1 +λ3 . Inserting this into (i) yields λ2 = ex +λ3 ≥ ex > 0. Because
λ2 > 0, (iv) implies that x = −1. So any solution must lie on the line (II) in Fig. A14.9.3
in the book. This implies that the third constraint y ≥ 1 must be slack.34 So from (v)
we get λ3 = 0, and then (ii) gives λ1 = ex+y − 1 ≥ e4 − 1 > 0. Thus from (iii), the first
constraint is active, so y = 4 − x. We already showed that x = −1, so y = 5.
34
Algebraically, because x + y ≥ 4 and x = −1, we have y ≥ 4 − x = 5 > 1.
71
(c) Hence the only possible solution is (x∗ , y ∗ ) = (−1, 5). Because L(x, y) is concave, as the
sum of concave functions, we have found the optimal point.
4. (a) The feasible set is shown in Fig. A14.9.4 in the book. The function to be maximized is
f (x, y) = x + ay. The level curves of this function are straight lines with slope −1/a if
a 6= 0, and vertical lines if a = 0. The dashed line in the figure is such a level curve (for
a ≈ −0.25). The maximum point for f is that point in the feasible region that we shall
find if we make a parallel displacement of this line as far to the right as possible without
losing contact with the shaded region.35
To apply the recipe of Section 14.9, we write the second constraint as −x − y ≤ 0, and
consider the Lagrangian L(x, y) = x + ay − λ1 (x2 + y 2 − 1) + λ2 (x + y). The Kuhn–Tucker
conditions are:
(iii) λ1 ≥ 0, with λ1 = 0 if x2 + y 2 < 1,
(i) L′1 (x, y) = 1 − 2λ1 x + λ2 = 0,
(ii) L′2 (x, y) = a − 2λ1 y + λ2 = 0,
(iv) λ2 ≥ 0, with λ2 = 0 if x + y > 0.
(b) From (i), 2λ1 x = 1 + λ2 ≥ 1. But (iii) implies that λ1 ≥ 0, so in fact λ1 > 0 and x > 0.
Because λ1 > 0, (iii) also implies that any maximum point must lie on the semi-circle
where x > 0 and x2 + y 2 = 1.
First consider the case x + y = 0. Because x2 + y 2 = 1 and x > 0, the only possible
√
√
solution is x = 21 2 and y = − 21 2.
Now, adding equations (i) and (ii) gives 0 = 1 + a − 2λ1 (x + y) + 2λ2 = 1 + a + 2λ2 since
x + y = 0. But λ2 ≥ 0, so a = 1 − 2λ2 ≤ −1 in this case. Then equation (i) along with
√
√
λ2 = 21 (1 − a) and x = 12 2 gives λ1 = (1 + λ2 )/2x = (1 − a)/4x = 2(1 − a)/4.
Second, consider the alternative case x + y > 0. Then (iv) implies that λ2 = 0, so (i) and
(ii) reduce to 1 − 2λ1 x = 0 and a − 2λ1 y = 0, and so x = 1/(2λ1 ) and y = a/(2λ1 ).
√
Inserting these into x2 +y 2 = 1 yields (1/4λ1 )2 (1+a2 ) = 1, and so λ1 = 21 1 + a2 . Hence
x = (1 + a2 )−1/2 and y = a(1 + a2 )−1/2 . Because x + y = (1 + a)(1 + a2 )−1/2 > 0 in this
case, it must occur when a > −1.
Conclusion: The only points satisfying the Kuhn–Tucker conditions are those given in
the text. Since the feasible set is closed and bounded and f is continuous, it follows from
the extreme value theorem that extreme points exists.
5. The Lagrangian is L = y − x2 + λy + µ(y − x + 2) − ν(y 2 − x). The Kuhn—Tucker conditions
are:
(i) L′x = −2x − µ + ν = 0;
(ii)
L′y
(iii) λ ≥ 0, with λ = 0 if y > 0;
= 1 + λ + µ − 2νy = 0;
(iv) µ ≥ 0, with µ = 0 if y − x > −2;
(v) ν ≥ 0, with ν = 0 if y 2 < x.
From (ii), 2νy = 1 + λ + µ > 0, so y > 0 and ν > 0, implying from (v) that y 2 = x. Also (iii)
implies λ = 0, and 2νy = 1 + µ. Also, (i) implies that x = 12 (ν − µ).
35
Why to the right?
72
Suppose µ > 0. Then y − x + 2 = y − y 2 + 2 = 0 with roots y = −1 and y = 2. Only y = 2
is feasible. Then x = y 2 = 4. Because λ = 0, conditions (i) and (ii) become −µ + ν = 8 and
µ − 4ν = −1, so ν = −7/3. This contradicts ν ≥ 0, so (x, y) = (4, 2) is not a candidate.
Therefore µ = 0 after all. Thus x = 12 ν = y 2 and, by (ii), 1 = 2νy = 4y 3 . Hence y = 4−1/3 ,
implying x = 4−2/3 . This is the only remaining candidate. It is the solution with λ = 0,
µ = 0, and ν = 1/2y = 4−1/6 .
6. (a) See Fig. A14.9.6 in the book. Note that for (x, y) to be admissible, one must have
e−x ≤ y ≤ 2/3, and so ex ≥ 3/2. This implies, in particular, that x > 0.
(b) The Lagrangian is L = −(x + 12 )2 − 12 y 2 − λ1 (e−x − y) − λ2 (y − 23 ), so the Kuhn–Tucker
conditions are:
(i) L′1 (x, y) = −(2x + 1) + λ1 e−x = 0;
(ii) L′2 (x, y) = −y + λ1 − λ2 = 0;
(iii) λ1 ≥ 0, with λ1 = 0 if e−x < y;
(iv) λ2 ≥ 0, with λ2 = 0 if y < 2/3.
From (i), λ1 = (2x + 1)ex ≥ 3/2, because of part (a), implying that y = e−x . From
(ii), λ2 = λ1 − y ≥ 3/2 − 2/3 > 0, so y = 2/3 because of (iv). This gives the solution
candidate (x∗ , y ∗ ) = (ln(3/2), 2/3), with λ1 = 3[ln(3/2) + 1/2] and λ2 = 3 ln(3/2) + 5/6.
The Lagrangian is easily seen to be concave as a function of (x, y) when λ1 ≥ 0, so this
is indeed the solution.
Here is an alternative argument: Suppose λ1 = 0. Then from (ii), y = −λ2 ≤ 0,
contradicting y ≥ e−x . So λ1 > 0, and (iii) gives y = e−x .
Suppose λ2 = 0. Then from (ii), λ1 = y = e−x and (i) gives e−2x = 2x + 1. Define
g(x) = 2x + 1 − e−2x . Then g(0) = 0 and g ′ (x) = 2 + 2e−2x > 0. So the equation
e−2x = 2x + 1 has no solution except x = 0. Thus λ2 > 0, etc.
14.10 Nonnegativity constraints
2. The Lagrangian, without multipliers for the non-negativity constraints, is L = xey−x − 2ey −
λ(y − 1 − x/2), so the first-order conditions (14.10.3) and (14.10.4) are
(i) L′x = ey−x − xey−x + 12 λ ≤ 0, and L′x = 0 if x > 0;
(ii) L′y = xey−x − 2e − λ ≤ 0, and L′y = 0 if y > 0;
(iii) λ ≥ 0, with λ = 0 if y < 1 + 21 x.
If x = 0, then (i) implies ey + 21 λ ≤ 0, which is impossible, so x > 0. Then from (i) we get
(iv) xey−x = ey−x + 12 λ.
Suppose first that λ > 0. Then (iii) and y ≤ 1 + x/2 imply (v) y = 1 + 12 x. Thus y > 0 and
1
from (ii) we have xey−x = 2e + λ. Using (iv) and (v), we get λ = 2ey−x − 4e = 2e(e− 2 x − 2).
1
But then λ > 0 implies that e− 2 x > 2, which contradicts x ≥ 0.
This leaves λ = 0 as the only possibility. Then (iv) gives x = 1. If y > 0, then (ii) yields
ey−1 = 2e, and so y − 1 = ln(2e) = ln 2 + 1. With x = 1 this contradicts the constraint
73
y ≤ 1 + 12 x. Hence y = 0, so we see that (x, y) = (1, 0) is the only point satisfying all the
first-order conditions, with λ = 0.36
3. The Lagrangian is L = x1 + 3x2 − x21 − x22 − k 2 − λ(x1 − k) − µ(x2 − k). A feasible triple
(x∗1 , x∗2 , k ∗ ) solves the problem if and only if there exist numbers λ and µ such that
(i) ∂L/∂x1 = 1 − 2x∗1 − λ ≤ 0, and 1 − 2x∗1 − λ = 0 if x∗1 > 0;
(ii) ∂L/∂x2 = 3 − 2x∗2 − µ ≤ 0, and 3 − 2x∗2 − µ = 0 if x∗2 > 0;
(iii) ∂L/∂k = −2k ∗ + λ + µ ≤ 0, and −2k ∗ + λ + µ = 0 if k ∗ > 0;
(iv) λ ≥ 0, with λ = 0 if x∗1 < k ∗ ;
(v) µ ≥ 0, with µ = 0 if x∗2 < k ∗ .
If k ∗ = 0, then feasibility requires x∗1 = 0 and x∗2 = 0, and so (i) and (ii) imply that λ ≥ 1
and µ ≥ 3, which contradicts (iii). Thus, k ∗ > 0. Next, if µ = 0, then (ii) and (iii) imply
that x∗2 ≥ 3/2 and λ = 2k ∗ > 0. So x∗1 = k ∗ = 1/4, contradicting x∗2 ≤ k ∗ . So µ > 0, which
implies that x∗2 = k ∗ . Now, if x∗1 = 0 < k ∗ , then λ = 0, which contradicts (i).
So 0 < x∗1 = 12 (1 − λ). Next, if λ > 0, then x∗1 = k ∗ = x∗2 = 21 (1 − λ) = 12 (3 − µ) =
1
2 (λ + µ) by (i), (ii), and (iii) respectively. But the last two equalities are only satisfied when
λ = −1/3 and µ = 5/3, which contradicts λ ≥ 0. So λ = 0 after all, with x∗2 = k ∗ > 0,
µ > 0, x∗1 = 12 (1 − λ) = 12 . Now, from (iii) it follows that µ = 2k ∗ and so, from (ii), that
3 = 2x∗2 + µ = 4k ∗ . The only possible solution is, therefore, (x∗1 , x∗2 , k ∗ ) = (1/2, 3/4, 3/4),
with λ = 0 and µ = 3/2.
Finally, with λ = 0 and µ = 32 , the Lagrangian x1 + 3x2 − x21 − x22 − k 2 − 32 (x2 − k) is a
quadratic function of (x1 , x2 , k), which has a maximum at the critical point (x∗1 , x∗2 , k ∗ ). As
stated at the end of the recipe in Section 14.9, this is sufficient for the same (x∗1 , x∗2 , k ∗ ) to
solve the problem.
Review exercises for Chapter 14
3. (a) If sales x of the first commodity are increased, the increase in net profit per unit increase
in x is the sum of three terms:
(i) p(x∗ ), which is the gain in revenue due to the extra output;
(ii) −p′ (x∗ )x∗ , which is the loss in revenue from selling x∗ units due to the reduced
price;
(iii) −C1′ (x∗ , y ∗ ), which is minus the marginal cost of the additional output.
In fact p(x∗ ) + p′ (x∗ )x∗ is the derivative of the revenue function R(x) = p(x)x at x = x∗ ,
usually called the marginal revenue. The first necessary condition therefore states that
the marginal revenue from increasing x must equal the (partial) marginal cost. The
argument when considering any variation in the sales y of the second commodity is just
the same.
36
The extreme value theorem cannot be applied here because the feasible set (the set of all points satisfying the
constraints) is unbounded — it includes the point (x, 0) for arbitrarily large x. However, you were told to assume
that the problem has a solution.
74
(b) With the restriction x + y ≤ m, we have to add the condition λ ≥ 0, with λ = 0 if
x̂ + ŷ < m.
5. (a) The Lagrangian is L = x2 + y 2 − 2x + 1 − λ( 14 x2 + y 2 − b); the first-order conditions are:
1
L′1 = 2x − 2 − λx = 0;
2
L′2 = 2y − 2λy = 0;
1 2
x + y 2 = b.
4
From (ii), one has (1 − λ)y = 0, and thus λ = 1 or y = 0.
(I) Suppose first that λ = 1. Then (i) gives x = 43 , and from (iii) we have y 2 =
q
b − 41 x2 = b − 94 , which gives y = ± b − 49 . This gives two candidates: (x1 , y1 ) =
q
q
(4/3, b − 94 ) and (x2 , y2 ) = (4/3, − b − 94 ).
√
(II) If y = 0, then√from (iii), x2 = 4b, i.e. x√= ±2 b. This gives two further candidates:
(x3 , y3 ) = (2 b, 0) and (x4 , y4 ) = (−2 b, 0).
Evaluating the objective function at these different solutions candidates gives:
f (x1 , y1 ) = f (x2 , y2 ) = b − 1/3;
√
√
f (x3 , y3 ) = (2 b − 1)2 = 4b − 4 b + 1;
√
√
f (x4 , y4 ) = (−2 b − 1)2 = 4b + 4 b + 1.
Clearly, (x4 , y4 ) is the maximum point. To decide which of the points (x3 , y3√
), (x1 , y1 ),
or (x2 , y2 ) give the minimum, we have to decide which of the two values 4b − 4 b + 1 and
b − 13 is smaller. The difference between these two is
√
2
√
√
4b − 4 b + 1 − b − 13 = 3 b − 43 b + 49 = 3
b − 23 > 0
since b > 49 . Thus the minimum occurs at (x1 , y1 ) and (x2 , y2 ).
y
4
(x1 , y1 )
2
(x4 , y4 )
-4
-2
2
-2
(x3 , y3 )
4
6
x
(x2 , y2 )
-4
Figure SM14.R.5
The constraint x2 /4 + y 2 = b describes the ellipse which is indicated by the solid curve
in Fig. SM14.R.5. The objective function f (x, y) = (x − 1)2 + y 2 is the square of the
75
distance between (x, y) and the point (1, 0). The level curves for f are therefore circles
centred at (1, 0), and in the figure we see those two that pass through the maximum and
minimum points.
√
√
7. (a) With L = x2 − 2x + 1 + y 2 − 2y − λ[(x + y) x + y + b − 2 a], the first-order conditions
are:
√
√
(i) L′1 = 2x − 2 − λ[ x + y + b + (x + y)/2 x + y + b] = 0,
√
√
(ii) L′2 = 2y − 2 − λ[ x + y + b + (x + y)/2 x + y + b] = 0.
From these equations it follows immediately that 2x−2 = 2y−2, so x = y. The constraint
√
√
gives 2x 2x + b = 2 a. Cancelling 2 and then squaring each side, one obtains the second
equation in (∗) of the main text.
(b) Differentiating yields:
dx = dy;
6x2 dx + x2 db + 2bx dx = da.
From these equations we easily read off the first-order partials of x and y w.r.t. a and b.
Further,
∂
∂2x
=
2
∂a
∂a
∂x
∂a
=
∂
12x + 2b ∂x
12x + 2b
6x + b
1
=−
=−
=− 3
.
2
2
2
2
3
∂a 6x + 2bx
(6x + 2bx) ∂a
(6x + 2bx)
4x (3x + b)3
8. With L = 10 − (x − 2)2 − (y − 1)2 − λ(x2 + y 2 − a), the Kuhn–Tucker conditions are:
(iii) λ ≥ 0, with λ = 0 if x2 + y 2 < a;
(i) L′x = −2(x − 2) − 2λx = 0;
(ii) L′y = −2(y − 1) − 2λy = 0.
Since the Lagrangian is concave when λ ≥ 0, these conditions are sufficient for a maximum.
One case occurs when λ = 0, implying that (x, y) = (2, 1). This is valid when a ≥ x2 +y 2 = 5.
The other case is when λ > 0. Then (i) implies that x = 2/(1 + λ) and (ii) implies that
2
y =p
1/(1 + λ). Because (iii) implies that x2 + y 2 = a, we have 5/(1 + λ)
p = apand so
λ = 5/a − 1, which is positive when a < 5. The solution then is (x, y) = (2 a/5, a/5).
9. (a) See the answer given in the book.
(b) The numbers (i)–(vi) in the following refer to the answer to (a) in the book. From (ii)
and (vi) we see that λ1 = 0 is impossible. Thus λ1 > 0, and from (iii) and (v), we see
that:
(x∗ )2 + r(y ∗ )2 = m.
(vii)
Now,
(I) Assume λ2 = 0. Then from (i) and (ii), y ∗ = 2λ1 x∗ and x∗ = 2λ1 ry ∗ , so x∗ =
√
r.
4λ21 rx∗ . But (vi) implies that x∗ 6= 0. Hence λ21 = 1/4r and thus λ1 = 1/2
p
√
∗
∗
∗
∗
Then y = x / pr, which inserted into (vii) and solved
for x yields x = m/2
p
∗
∗
m/2p
≥ 1 ⇐⇒ m ≥ 2. Thus,
and then y = m/2r. Note that x ≥
p1 ⇐⇒
√
for m ≥ 2, a solution candidate is x∗ = m/2 and y ∗ = m/2r, with λ1 = 1/2 r
and λ2 = 0.
76
(II) Assume λ2 > 0. Then x∗ = p
1 and from (vii) we have r(y ∗ )2 = m − 1. By
(ii), one has y ∗ ≥ 0, so y ∗ = (m − 1)/r. Inserting
p these values into (i) and
(ii), then solving
p for λ1 and λ2 , one obtains λ1 = 1/2 r(m − 1) and furthermore,
λ2 = (2−m)/ r(m − 1). Note that λ2 > 0 p
⇐⇒ 1 < m < 2. Thus, for p
1 < m < 2,
∗
∗
the only solution candidate
is x = 1, y = (m − 1)/r, with λ1 = 1/2 r(m − 1)
p
and λ2 = (2 − m)/ r(m − 1).
The objective function is continuous and the constraint set is obviously closed and
bounded, so by the extreme value theorem there has to be a maximum. The solution
candidates we have found are therefore optimal.37
√
√
√
(c) For m ≥ 2, one has V (r, m) = m/2 r, so Vm′ =√1/2 r = λ1 , and Vr′ = −m/4 r3 ,
√
whereas L′r = −λ1 (y ∗ )2 = −(1/2 r)m/2r = −m/4 r3 .
p
p
′ = 1/2 r(m − 1) = λ , and
For 1 < m <
2,
one
has
V
(r,
m)
=
(m
−
1)/r,
so
V
1
m
p
Vr′ = −(1/2) (m − 1)/rp3 , whereas
p
L′r = −λ1 (y ∗ )2 = −[1/2 r(m − 1)](m − 1)/r = −(1/2) (m − 1)/r3 .
15 Matrix and Vector Algebra
15.1 Systems of linear equations
6. The equation system is:





0.158x

x−



x
0.712y
−
c = −95.05
− s + 0.158c =
34.30
y−s+
c =
0
=
93.53
Solve the first equation for y as a function of c. Insert this expression for y and x = 93.53
into the third equation. Solve it to get s as a function of c. Insert the results into the second
equation and solve for c, and then solve for y and s in turn. The answer is given in the main
text.
15.3 Matrix multiplication
6. We know that A is an m × n matrix. Let B be a p × q matrix. The matrix product AB is
defined if and only if n = p, and BA is defined if and only if q = m. So for both AB and
BA to be defined, it is necessary and sufficient that B is an n × m matrix.
!
x y
.
7. We know from the previous exercise that B must be a 2 × 2 matrix. So let B =
z w
!
!
!
x + 2y 2x + 3y
1 2
x y
=
Then BA =
z + 2w 2z + 3w
2 3
z w
37
Alternatively, L′′11 = −2λ1 ≤ 0, L′′22 = −2rλ1 ≤ 0, and ∆ = L′′11 L′′22 − (L′′12 )2 = 4rλ21 − 1. In the case m ≥ 2,
∆ = 0, and in the case 1 < m < 2, ∆ = 1/(m − 1) > 0. Thus in both cases, L(x, y) is concave. So the Kuhn–Tucker
conditions are sufficient for a maximum.
77
!
!
!
1 2
x y
x + 2z y + 2w
while AB =
=
.
2 3
z w
2x + 3z 2y + 3w
Hence, BA = AB if and only if:
(i) x + 2y = x + 2z; (ii) 2x + 3y = y + 2w; (iii) z + 2w = 2x + 3z; and (iv) 2z + 3w = 2y + 3w.
Equations (i) and (iv) are both true if and only if y = z; then equations (ii) and (iii) are also
both true if and only if, in addition, x = w − y. To summarize, all four equations hold if
and only if y = z and x = w − y. Hence,
the matrices
B that commute
with A are precisely
!
!
!
−1 1
1 0
w−y y
where y and w can be any
+y
=w
those of the form B =
1 0
0 1
y
w
real numbers.
15.4 Rules for matrix multiplication
2. We start by performing the multiplication

  

a d e
x
ax + dy + ez

  

d b f  y  = dx + by + f z  .
e f c
z
ex + f y + cz
Next,

ax + dy + ez


(x, y, z) dx + by + f z  = ax2 + by 2 + cz 2 + 2dxy + 2exz + 2f yz,
ex + f y + cz

which is a 1 × 1 matrix.
8. (a) Direct verification yields (i) because
2
A = (a + d)A − (ad − bc)I2 =
a2 + bc ab + bd
ac + cd bc + d2
!
(b) For the matrix A in part (a), one has A2 = 0 if a + d = 0 and ad = bc, so one example
with A2 = 0 6= A is
!
1
1
A=
.
−1 −1
(c) By part (a), A3 = (a+d)A2 −(ad−bc)A. So A3 = 0 implies that (a+d)A2 = (ad−bc)A
and then, multiplying each side by A once more, that (a+d)A3 = (ad−bc)A2 . If A3 = 0,
therefore, one has two cases:
(i) A2 = 0;
(ii) ad − bc = 0.
But in case (ii), one has (a + d)A2 = 0, which gives rise to two subcases:
(ii)(a) A2 = 0;
(ii)(b) ad − bc = 0 and a + d = 0.
Now, even in case (ii)(b), the result of part (a) implies that A2 = 0, which is therefore
true in every case.
78
15.5 The transpose
6. In general, for any natural number n > 3, one has
[(A1 A2 · · · An−1 )An ]′ = A′n [A1 A2 · · · An−1 ]′ .
As the induction hypothesis, suppose the result is true for n − 1. Then the last expression
becomes A′n A′n−1 · · · , A′2 A′1 , so the result is true for n.
8. (a) TS = S is shown in the book. A similar argument shows that T2 = 21 T + 12 S. To
prove the last equality, we do not have to consider the individual elements. Instead, we
premultiply the last equation by T and then use TS = S to obtain
T3 = TT2 = T
1
2T
+ 12 S = 21 T2 + 12 TS =
1
2
1
2T
(b) We prove by induction that the appropriate formula is
+ 12 S + 21 S = 14 T + 34 S.
Tn = 21−n T + (1 − 21−n )S.
(∗)
This formula is correct for n = 1.38 Suppose (∗) is true for n = k. Then premultiplying
by T and using the two first equalities in (a), one obtains
Tk+1 = TTk = T(21−k T + (1 − 21−k )S) = 21−k T2 + (1 − 21−k )TS =
21−k ( 21 T + 12 S) + (1 − 21−k )S = 2−k T + 2−k S + S − 2 · 2−k S = 2−k T + (1 − 2−k )S
which is formula (∗) for n = k + 1.
15.6 Gaussian elimination
3. Apply the following elementary operations successively:
(i) subtract the third equation from the first;
(ii) subtract the new first equation from the two others;
(iii) interchange the second and the third equation;
(iv) multiply the second equation by −3 and add it to the third equation.
The result is
w
x
y
2

1
1
1
3
1
4
2
2

z

 
1 0 2
2
1 − c2
3 1

 
−1 3c  ∼ 0 1 0 −1
2c2 − 1 
1 c2
0 0 0
0 −5c2 + 3c + 2
The last row of the last matrix indicates that the system has solutions if and only if −5c2 +
3c + 2 = 0, that is, if and only if c = 1 or c = −2/5. For these particular values of c we get
the solutions in the book.39
38
39
And, by part (a), also for n = 2, 3.
The final answer can take many equivalent forms depending on how you arrange the elementary operations.
79
4. After moving the first row down to row number three, then applying the indicated elementary
row operations, we obtain the successive augmented matrices




−3 −a
1 2
1
b2
1
2
1
b2




∼ 0
7
b3  ←
−2
4 b3 − 3b2  × − 12
3 4
a 1 a + 1 b1 ←
0 1 − 2a 1 b1 − ab2


1
2
1
b2


∼ 0
1
−2 32 b2 − 21 b3  2a − 1
←
0 1 − 2a 1
b1 − ab2


1 2
1
b2


1
3
∼ 0 1
−2

2 b2 − 2 b3
0 0 3 − 4a b1 + (2a − 23 )b2 + ( 21 − a)b3
Looking at the element in row 3 and column 3 of the final augmented matrix makes it obvious
that there is a unique solution if and only if 3 − 4a 6= 0 or equivalently a 6= 3/4.
5. Put a = 3/4 in Exercise 15.6.4. Then the last row in the last augmented matrix in the above
solution becomes (0, 0, 0, b1 − 14 b3 ). It follows that if b1 6= 41 b3 there is no solution. If b1 = 14 b3
there is an infinite set of solutions. For an arbitrary real t, there is a unique solution with
z = t. Then the second equation gives y = 23 b2 − 12 b3 + 2t, and finally the first equation gives
x = −2b2 + b3 − 5t.
15.8 Geometric interpretation of vectors
2. (a) See the answer given in the book.
(b) See Fig. A15.8.2 in the book. According to the point–point formula, the line through (3, 1)
and (−1, 2) has the equation x2 = − 41 x1 + 47 or x1 + 4x2 = 7. Call this line L. Any point
(x1 , x2 ) on L satisfies x1 + 4x2 = 7 and equals (−1 + 4λ, 2 − λ) for λ = 41 (x1 + 1) = 2 − x2 .
So there is a one-to-one correspondence between points:
(i) that lie on the line segment joining a = (3, 1) and b = (−1, 2);
(ii) whose coordinates can be written as (−1 + 4λ, 2 − λ) for some λ in [0, 1].
(c) This is, precisely, line L: as we let λ ∈ R, the previous argument gives us a one-to-one
correspondence between points:
(i) that lie on the line L going through a = (3, 1) and b = (−1, 2);
(ii) whose coordinates can be written as (−1 + 4λ, 2 − λ) for some λ in R.
15.9 Lines and planes
3. First, note that (5, 2, 1) − (1, 0, 2) = (4, 2, −1) and (2, −1, 4) − (1, 0, 2) = (1, −1, 2) are two
vectors in the plane. The normal (p1 , p2 , p3 ) to the plane must be orthogonal to both these
vectors, so (4, 2, −1) · (p1 , p2 , p3 ) = 4p1 + 2p2 − p3 = 0 and (1, −1, 2) · (p1 , p2 , p3 ) = p1 −
p2 + 2p3 = 0. One solution of these two equations is (p1 , p2 , p3 ) = (1, −3, −2). Then using
formula (4) with (a1 , a2 , a3 ) = (2, −1, 4) yields (1, −3, −2) · (x1 − 2, x2 + 1, x3 − 4) = 0, or
x1 − 3x2 − 2x3 = −3.40
40
A more pedestrian approach is to assume that the equation is ax + by + cz = d and require the three points to
satisfy the equation: a + 2c = d, 5a + 2b + c = d, 2a − b + 4c = d. Solve for a, b, and c in terms of d, insert the results
80
Review exercises for Chapter 15
7. (a) See the answer given in the book.
(b) By elementary operations:

2
1
3
2
−3
4


−1 2 ←
1
1 0 ← ∼ 2
−1 1
3

1
∼ 0
0

1

∼ 0
0

1
∼ 0
0


−2 −3
1 0
1 −3
∼ 0 8
−1 2 ←
−1 1 ←
0 13


−3
1
0
1 −3
1 −3/8 1/4 −13 ∼ 0 1
←
13
−4
1
0 0

←
−3
1
0

3
1 −3/8
1/4
0
1
−18/7


←
1
0 −1/8
3/4
∼ 0
1 −3/8
1/4  ←
3/8 1/8
0
0
1
−18/7
−3
2
4

1 0
−3 2 1/8
−4 1
1
−3/8
7/8

0
1/4 
−9/4 8/7
0 0
1 0
0 1

3/7
−5/7  .
−18/7
The solution is x1 = 3/7, x2 = −5/7, x3 = −18/7.
(c) We again use elementary operations:
!
!
−5
1
3
4
0
1 3 4 0
∼
0 −14 −19 0 −1/14
5 1 1 0 ←
!
1 3
4
0 ←
∼
∼
−3
0 1 19/14 0
1 0 −1/14 0
0 1 19/14 0
!
The solution is x1 = (1/14)x3 , x2 = −(19/14)x3 , for arbitrary x3 . That is, there is one
degree of freedom.
10. (a) See the answer given in the book.
(b) In part (a) we saw that a can be produced even without throwing away outputs.
For b to be possible if we are allowed to throw away output, there must exist a scalar λ
in [0, 1] such that 6λ + 2 ≥ 7, −2λ + 6 ≥ 5, and −6λ + 10 ≥ 5.
These inequalities reduce to λ ≥ 5/6, λ ≤ 1/2, λ ≤ 5/6, which are incompatible.
(c) Revenue is R(λ) = p1 x1 + p2 x2 + p3 x3 = (6p1 − 2p2 − 6p3 )λ + 2p1 + 6p2 + 10p3 .
If the constant slope 6p1 − 2p2 − 6p3 is > 0, then R(λ) is maximized at λ = 1;
if 6p1 − 2p2 − 6p3 is < 0, then R(λ) is maximized at λ = 0.
Only in the special case where 6p1 − 2p2 − 6p3 = 0 can the two plants both remain in use.
11. (a) If PQ − QP = P, then PQ = QP + P, and so
P2 Q = P(PQ) = P(QP + P) = (PQ)P + P2 = (QP + P)P + P2 = QP2 + 2P2 .
Thus, P2 Q − QP2 = 2P2 . Moreover,
P3 Q = P(P2 Q) = P(QP2 + 2P2 ) = (PQ)P2 + 2P3 = (QP + P)P2 + 2P3 = QP3 + 3P3 .
into the equation ax + by + cz = d and cancel d.
81
Hence, P3 Q − QP3 = 3P3 .
To prove the result for general k, suppose that Pn Q − QPn = nPn as the induction
hypothesis for n = k. Then, for n = k + 1, one has
Pn Q = P(Pk Q) = P(QPk + kPk ) = (PQ)Pk + kPk+1
= (QP + P)Pk + kPk+1 = QPk+1 + (k + 1)Pk+1 ,
so the induction hypothesis is also true for n = k + 1.
16 Determinants and Inverse Matrices
16.1 Determinants of order 2
9. (a) See the answer given in the book.
(b) The suggested substitutions produce the two equations
Y1 = c1 Y1 + A1 + m2 Y2 − m1 Y1 ;
Y2 = c2 Y2 + A2 + m1 Y1 − m2 Y2
or
(1 − c1 − m1 )Y1 − m2 Y2 = A1 ;
−m1 Y1 + (1 − c2 − m2 )Y2 = A2
This can be rewritten in the matrix form
1 − c 1 − m1
−m2
−m1
1 − c2 − m2
!
Y1
Y2
!
=
A1
A2
!
For these equations to be soluble, we need to assume that D = (1 − c1 − m1 )(1 − c2 −
m2 ) − m1 m2 6= 0. When D 6= 0, the answers in the text can be derived using Cramer’s
rule.
(c) Y2 depends linearly on A1 . Economists usually assume that D given in part (b) is positive,
as it will be provided that the parameters c1 , c2 , m1 , m2 are all sufficiently small. Then
increasing A1 by one unit changes Y2 by the factor m1 /D ≥ 0, so Y2 increases when A1
increases.
Here is an economic explanation: An increase in A1 increases nation 1’s income, Y1 . This
in turn increases nation 1’s imports, M1 . However, nation 1’s imports are nation 2’s
exports, so this causes nation 2’s income, Y2 , to increase, and so on.
16.2 Determinants of order 3
1 −1 0
1. (a) Sarrus’s rule yields 1 3 2 = 0 − 2 + 0 − 0 − 0 − 0 = −2.
1 0 0
1 −1 0
(b) By Sarrus’s rule 1
3 2 = 3 − 2 − 0 − 0 − 4 − (−1) = −2.
1
2 1
82
(c) Because a21 = a31 = a32 = 0, the only non-zero term in the expansion (16.2.2) is the
product of the terms on the main diagonal. The determinant is therefore adf .41
a 0 b
(d) By Sarrus’s rule, 0 e 0 = aed + 0 + 0 − bec − 0 − 0 = e(ad − bc).
c 0 d
1 −1
1
3. (a) The determinant of the coefficient matrix is |A| = 1
1 −1 = −4.
−1 −1 −1
The numerators in (16.2.6) are
2 −1
1
0
1 −1 = −4 ,
−6 −1 −1
1
2
1
1
0 −1 = −8 ,
−1 −6 −1
1 −1
2
1
1
0 = −12
−1 −1 −6
Hence, (16.2.6) yields the solution x1 = 1, x2 = 2, and x3 = 3. Inserting this into the
original system of equations confirms that this is a correct answer.
(b) The determinant of the coefficient matrix is equal to −2, and the numerators in (16.2.6)
are all 0, so the unique solution is x1 = x2 = x3 = 0.
(c) Follow the pattern in part (a) to get the answer in the book.
8. (a) Substituting T = d+tY into the expression for C gives C = a−bd+b(1−t)Y . Substituting
for C in the expression for Y then yields Y = a + b(Y − d − tY ) + A0 . Then solve for Y ,
T , and C in turn to derive the answers given in (b) below.
   

A0
Y
1 −1 0
   

(b) We write the system as −b 1 b  C  =  a  .
d
T
−t 0 1
1 −1 0
With D = −b 1 b = 1 + bt − b, Cramer’s rule yields42
−t 0 1
A0 −1 0
a − bd + A0
1
,
Y =
a
1 b =
D
1 − b(1 − t)
d
0 1
and T =
1 A0 0
1
a − bd + A0 b(1 − t)
C=
,
−b a b =
D
1 − b(1 − t)
−t d 1
1 −1 A0
1
t(a + A0 ) + (1 − b)d
.
−b 1
a =
D
1 − b(1 − t)
−t 0
d
16.3 Determinants in general
1. (a) Each of the three determinants is a sum of 4! = 24 terms. In (a) there is only one nonzero
term. In fact, according to (16.3.4), the value of the determinant is 24.
41
Alternatively, Sarrus’s rule gives the same answer.
This problem is meant to train you in using Cramer’s rule. It is also a warning against its overuse, since solving
the equations by systematic elimination is much more efficient.
42
83
(b) Only two terms in the sum are nonzero: the product of the elements on the main diagonal,
which is 1 · 1 · 1 · d, with a plus sign; and the term
1
0
0
a
0
1
0
b
0
0
1
c
1
0
0
d
Since there are 5 rising lines between the pairs, the sign of the product 1 · 1 · 1 · a must be
minus. So the value of the determinant is d − a. (c) 4 terms are nonzero. See the answer
given in the book.
16.4 Basic rules for determinants
14. The answer in the text amounts to the following successive transformations
a+b
a
···
a
a + b ···
Dn =
..
..
..
.
.
.
a
a
···
a
na + b na + b · · ·
a
a
a + b ···
.. =
..
..
..
.
.
.
.
a
a
a
···
1
1
···
a a + b ···
= (na + b) .
..
..
..
.
.
a
a
···
na + b
a
..
.
a+b
1 1 ···
0 b ···
= (na + b) . . .
.. ..
..
0 0 ···
a+b
1
a
..
.
1
0
..
.
b
According to (16.3.4), the last determinant is bn−1 . Thus Dn = (na + b)bn−1 .
16.5 Expansion by cofactors
1. (a) See the answer given in the book.
(b) One possibility is to expand by the second row or the third column, because they both
have two zero entries. But it is easier first to use elementary operations to get a row or
a column with at most one non-zero element. For example:
−2
1
2
3 4
0 −1
0 11
2 −1
0 3 ←
−2
0 −1 3 ←
2
−5
−1
0
11
= 0 −6 −60 ←
0
5
55 ←
4
1
2
3
4
−1
0 11
0 −1
0 11
=
= −5 −6 −5
0 −5 −6 −5
4
5 11
0
4
5 11
= −1
−6 −60
= −(−330 + 300) = 30
5
55
(c) See the book for the simple answer. When computing determinants one can use elementary column as well as row operations, but column operations become meaningless when
solving linear equation systems using Gaussian elimination.
84
16.6 The inverse of a matrix
!
!
!
−1/2
5
1
0
3/2
−5
= I.
+
=
8. By direct computation, B2 + B =
0 1
1/4 −1/2
−1/4 3/2
One can either verify by direct matrix multiplication that B3 − 2B + I = 0, or somewhat
more easily, use the relation B2 + B = I to argue that B2 = I − B and so
B3 − 2B + I = B(I − B) − 2B + I = B − B2 − 2B + I = −B2 − B + I = 0.
Furthermore B2 +B = I implies that B(B+I) = I. It follows from (16.6.4) that B−1 = B+I.
16.7 A general formula for the inverse
1. (a) |A| = 10 − 12 = −2, and the adjugate is
A−1
1
=−
2
C11 C21
C12 C22
5 −3
−4
2
!
=
!
=
!
5 −3
, so the inverse is
−4
2
−5/2 3/2
2
−1
!

 
1
4
2
C11 C21 C31


 
(b) The adjugate of B is adj B = C12 C22 C32  = 2 −1
4
4 −2 −1
C13 C23 C33
and |B| = b11 C11 + b21 C21 + b31 C31 = 1 · 1 + 2 · 4 + 0 · 2 = 9 by expansion along the first
column. Hence,


1
4
2
1
1

B−1 = (adj B) = 2 −1
4
9
9
4 −2 −1

(c) Since the second column of C equals −2 times its third column, the determinant of C is
zero, so there is no inverse.
3. The determinant of I − A is |I − A| = 0.496, and

0.72

adj (I − A) = 0.08
0.16
the adjugate is

0.64 0.40

0.76 0.32 .
0.28 0.64


1.45161 1.29032 0.80645
1


Hence (I − A)−1 =
adj(I − A) ≈ 0.16129 1.53226 0.64516 ,
0.496
0.32258 0.56452 1.29032
rounded to five decimal
places.
For
an
exact
answer, note
that 1000/496 = 125/62



18 16 10
0.72 0.64 0.40
1 



and adj(I − A) = 0.08 0.76 0.32 =
 2 19 8,
25
4 7 16
0.16 0.28 0.64


18 16 10
5 

which gives (I − A)−1 =
 2 19 8 .
62
4 7 16
85
4. Let B denote the n × p matrix whose k th column has the elements b1k , b2k , . . . , bnk . The p
systems of n equations in n unknowns can be expressed as AX = B, where A is n × n and
X is n × p. Following the method illustrated in Example 2, exactly the same row operations
that transform the n × 2n matrix (A : I) into (I : A−1 ) will also transform the n × (n + p)
matrix (A : B) into (I : B∗ ), where B∗ is the matrix with elements b∗ij .43 When k = r, the
solution to the system is x1 = b∗1r , x2 = b∗2r , . . . , xn = b∗nr .
5. (a) Using elementary operations,
1 2
3 4
∼
1 2
0 1
1
3
2
1 0
0 1
!
0
− 12
!
−3
∼
←
1
2
0 −2
←
∼
−2
1 0
0 1
1 0
−3 1
−2
3
2
1
− 21
!
!
− 12
(b) In this case,

1 2
2 4
3 5

1
2
∼ 0 −1
0
0

1 0 −3
∼ 0 1
3
0 0
1


−2 −3
1 0 0
1
∼ 0
0 1 0 ←
0 0 1 ←
0


1
3
1 0 0
−3
−3 0 1 ×(−1) ∼ 0
−2 1 0 ×(−1)
0
−1


1
−5
0
2 ←
∼ 0
3
0 −1 ←
−3 3
2 −1
0
0
3
5
6
2
3
0 −1
−1 −3
2 3
1 3
0 1
0 0
1 0
0 1

1 0 0
−2 1 0 ←
−3 0 1 ←

1
0
0 ←
3
0 −1 −2
2 −1
0

1 −3
2
−3
3 −1
2 −1
0
(c) We see that the third row equals the first row multiplied by −3, so the matrix has no
inverse.
16.8 Cramer’s Rule
1. (a) The determinant |A| of the coefficient matrix is
1
2 −1
|A| = 2 −1
1 = 19.
1 −1 −3
The determinants in (16.8.2) are
−5
2 −1
6 −1
1 = 19,
−3 −1 −3
1 −5 −1
2
6
1 = −38,
1 −3 −3
1
2 −5
2 −1
6 = 38
1 −1 −3
According to (16.8.4) the solution is
x = 19/19 = 1,
y = −38/19 = −2,
and
z = 38/19 = 2
Inserting this into the original system of equations confirms that this is the correct answer.
43
In fact, because these row operations are together equivalent to premultiplication by A−1 , it must be true that
B = A−1 B.
∗
86
1 1 0 0
1 0 1 0
. Subtracting the fourth
(b) The determinant |A| of the coefficient matrix is
0 1 1 1
0 1 0 1
column from the second leaves only one non-zero element in the second column, and so
1 1 0
reduces the determinant to − 0 1 1 = −1.
0 0 1
We ask you to check that the other determinants in (16.8.2) are
3
2
6
1
1
0
1
1
0
1
1
0
0
0
= 3,
1
1
1
1
0
0
3
2
6
1
0
1
1
0
0
0
= −6 ,
1
1
1
1
0
0
1
0
1
1
3
2
6
1
0
0
= −5 ,
1
1
1
1
0
0
1
0
1
1
0
1
1
0
2
3
= 5.
6
1
According to (16.8.4) the solution is x = −3, y = 6, z = 5, and u = −5. Inserting this
into the original system of equations confirms that this is the correct answer.44
3. According to Theorem 16.8.2, the system has nontrivial solutions if and only if the determinant of the coefficient equal to 0. Expansion along the first row gives
a b c
b c
b a
c a
= a(bc − a2 ) − b(b2 − ac) + c(ab − c2 )
+c
−b
b c a =a
c a
c b
a b
c a b
= 3abc − a3 − b3 − c3 .
Thus the system has nontrivial solutions if and only if 3abc − a3 − b3 − c3 = 0.
Review exercises for Chapter 16
5. One of many ways to evaluate the determinant is to expand along column 3, which gives
q −1 q − 2
q −1
1 −p
− (2 − p)
|A| = 1 −p 2 − p = (q − 2)
2 −1
2 −1
2 −1
0
= (q − 2)(−1 + 2p) − (2 − p)(−q + 2) = (q − 2)(p + 1),
q+1
0
q−1
q+1 q−1
= 2(p − 1)(q − 2).
Now, |A + E| = 2
1 − p 3 − p = (1 − p)
3
1
3
0
1
For the rest, see the answer in the text.
44
Of course, there is a much quicker way to solve these four equations. Subtracting the fourth from the third
yields z = 5 immediately. Then the second equation gives x = −3; the first gives y = 6; and the last gives u = −5.
87
8. (a) This becomes easy after noting that

 

n n ... n
1 1 ... 1
1 1 ... 1
1 1 . . . 1 1 1 . . . 1 n n . . . n

 


U2 =  . . .
. =  . .
.  = nU
. . .
 .. .. . . ..   .. .. . . . ..   .. .. . . . .. 
1 1 ... 1
1 1 ... 1
n n ... n

(b) The trick is to note that

 
 

3 3 3
1 0 0
4 3 3

 
 

A = 3 4 3 = 0 1 0 + 3 3 3 = I3 + 3U
3 3 3
0 0 1
3 3 4
From (a), (I3 + 3U)(I3 + bU) = I3 + (3 + b + 3 · 3bU) = I3 + (3 + 10b)U. This can be
made equal to I3 by choosing b = −3/10. It follows that
 




3
3
3
1 0 0
7 −3 −3
10
10
10
1 
 3


3
3 
A−1 = (I3 +3U)−1 = I3 −(3/10)U = 0 1 0− 10
7 −3
10
10  = 10 −3
3
3
3
0 0 1
−3 −3
7
10
10
10
10. (a) Gaussian elimination with elementary row operations



a 1 4 2 ←
1 0



2
∼ 2 1
2
2 1 a
1 0 −3 a ←
a 1



1 0
−3
a
1 0



2
∼ 0 1 a + 6 −2a + 2 −1 ∼ 0 1
0 1 3a + 4 −a2 + 2 ←
0 0
yields

−2
−3 a

2
a
2 ←
4 2 ←
−a

−3
a

a2 + 6
−2a + 2 
−a2 + 3a − 2 −a2 + 2a
It follows that the system has a unique solution if and only if −a2 + 3a − 2 6= 0 — i.e.,
if and only if a 6= 1 and a 6= 2. If a = 2, the last row consists only of 0’s so there are
infinitely many solutions, whereas if a = 1, there are no solutions.
(b) If we perform the same elementary operations on the associated extended matrix as in
part (a), then the fourth column is transformed as follows:
 
 




b1
b3
b3
b3
 
 




b2 − 2b3
b2  → b2  → b2 − 2b3  → 
.
b3
b1
b1 − ab3
b1 − b2 + (2 − a)b3


1 0
−3
b3


Thus, the final extended matrix is 0 1
a2 + 6
b2 − 2b3
.
2
0 0 −a + 3a − 2 b1 − b2 + (2 − a)b3
So there are infinitely many solutions if and only if all elements in the last row are 0,
which is true if and only if: either (i) a = 1 and b1 − b2 + b3 = 0; or (ii) a = 2 and b1 = b2 .
15. (a) See the answer given in the book.
88
(b) The trick is to note that the cofactor expansions of |A|, |B| and |C| along the rth row
P
P
P
take the respective forms nj=1 arj Crj , nj=1 brj Crj and nj=1 (arj + brj )Crj for exactly
the same collection of cofactors Crj , for j = 1, 2, . . . , n. Then,
|C| =
n
X
(arj + brj )Crj =
j=1
n
X
arj Crj +
j=1
n
X
j=1
brj Crj = |A| + |B|
16. It is a bad idea to use “brute force” here. Note instead that rows 1 and 3 in the determinant
have “much in common” and so do rows 2 and 4. So begin by subtracting row 3 from row 1,
and row 4 from row 2. According to Theorem 16.4.1(vi), this does not change the value of
the determinant. This gives, if we thereafter use Theorem 16.4.1(iii),
0
a−b
0
b−a
b−a
0
a−b
0
= (a − b)2
x
b
x
a
a
x
b
x
0
−1
x
a
1
0
b
x
0 −1
1 0
= (a − b)2
x a
b x
0
−1
x
a
1
0
b
x
0
0
1
0
x a+b
b
2x
The last equality is obtained by adding column 2 to column 4 in the middle determinant. If
we expand the last determinant along row 1, we obtain successively
−1 1
0
−(a − b) x x a + b = −(a − b)2 [−2x2 + b(a + b) − 2x2 + a(a + b)]
a b
2x
2
= (a − b)2 [4x2 − (a + b)2 ] = (a − b)2 [2x − (a + b)][2x + (a + b)].
The conclusion follows.
17 Linear Programming
17.1 A graphical approach
3. The set A corresponds to the shaded polygon in Fig. SM17.1.3. The following arguments
explain the answers given in the book.
(a) The solution is obviously at the uppermost point P in the polygon because it has the
largest x2 coordinate among all points in A. Point P is where the two lines −2x1 + x2 =
2 and x1 + 2x2 = 8 intersect, and the solution of these two equations is (x1 , x2 ) =
(4/5, 18/5).
(b) The point in A with the largest x1 coordinate is obviously Q = (8, 0).
(c) The line 3x1 + 2x2 = c for one typical value of c is the dashed line in Fig. SM17.1.3. As c
increases, the line moves out farther and farther to the north-east. The line that has the
largest value of c, and still has a point in common with A, is the one that passes through
the point Q in the figure.
(d) The line 2x1 − 2x2 = c (or x2 = x1 − c/2) makes a 45◦ angle with the x1 axis, and
intersects the x1 axis at c/2. As c decreases, the line moves up and to the left. The line
in this family that has the smallest value of c, and still has a point in common with A, is
the one that passes through the point P in the figure.
89
x2
−2x1 + x2 = 2
4
3x1 + 2x2 = c
P
A
Q
x1
8
x1 + 2x2 = 8
Figure SM17.1.3
(e) The line 2x1 + 4x2 = c is parallel to the line x1 + 2x2 = 8 in the figure. As c increases,
the line moves out farther and farther to the north-east. The line with points in common
with A that has the largest value of c is obviously the one that coincides with the line
x1 + 2x2 = 8. So all points on the line segment between P and Q are solutions.
(f) The line −3x1 − 2x2 = c is parallel to the dashed line in the figure, and intersects the x1
axis at −c/3. As c decreases, the line moves out farther and farther to the north-east, so
the solution is at Q = (8, 0). (We could also argue like this: Minimizing −3x1 −2x2 subject
to (x1 , x2 ) ∈ A is obviously equivalent to maximizing 3x1 + 2x2 subject to (x1 , x2 ) ∈ A,
so the solution is the same as the one in part (c).)
17.2 Introduction to Duality Theory
1. (a) See Fig. A17.1.1a in the answer section of the book. When 3x1 + 2x2 ≤ 6 is replaced by
3x1 + 2x2 ≤ 7 in Exercise 17.1.1, the feasible set expands because the steeper line through
P is moved out to the right. The new optimal point is at the intersection of the lines
3x1 + 2x2 = 7 and x1 + 4x2 = 4, and it follows that the solution is (x1 , x2 ) = (2, 1/2).
The old maximum value of the objective function was 36/5. The new optimal value is
3 · 2 + 4 · 21 = 8 = 40/5, and the difference in optimal value is u∗1 = 4/5.
(b) When x1 + 4x2 ≤ 4 is replaced by x1 + 4x2 ≤ 5, the feasible set expands because the
line x1 + 4x2 = 4 is moved up. The new optimal point is at the intersection of the lines
3x1 + 2x2 = 6 and x1 + 4x2 = 5, and it follows that the solution is (x1 , x2 ) = (7/5, 9/10).
The old maximum value of the objective function was 36/5. The new optimal value is
39/5, and the difference in optimal value is u∗2 = 3/5.
(c) See the answer given in the book.
90
17.3 The Duality Theorem
1. (a) From Fig. A17.3.1a in the text it is clear that as c increases, the dashed line moves out
farther and farther to the north-east. The line that has the largest value of c and still has
a point in common with the feasible set, is the one that passes through the point P , which
has coordinates (x, y) = (0, 3), where the associated maximum value is 2 · 0 + 3 · 7 = 21.
(b) In Fig. A17.3.1b, as c decreases, the dashed line moves farther and farther to the southwest. The line that has the smallest value of c and still has a point in common with the
feasible set, is the one that passes through the point P , which has coordinates (u1 , u2 ) =
(0, 1). The associated minimum value is 20u1 + 21u2 = 21.
(c) Yes, because the maximum of the primal in (a) and the minimum of the dual in (b) both
equal 21.
3. (a) See the answer given in the book.
(b) The problem is illustrated graphically in Fig. SM17.3.3, which makes clear the answer
given in the book.
x2
2x1 + x2 = 16
x1 + 2x2 = 11
5
(x1∗ , x2∗ ) = (7, 2)
x1 + 4x2 = 16
x1
5
10
400x1 + 500x2 = constant
Figure SM17.3.3
(c) Relaxing the first constraint to 2x1 + x2 ≤ 17 allows the solution to move out to the
intersection of the two lines 2x1 + x2 = 17 and x1 + 2x2 = 11. So the new solution is
x1 = 23/3, x2 = 5/3, where the profit is 3900.
Relaxing the second constraint to x1 + 4x2 ≤ 17 makes no difference, because some
capacity in division 2 remained unused anyway at (7, 2).
Relaxing the third constraint to x1 + 2x2 ≤ 12, the solution is no longer where the
lines 2x1 + x2 = 16 and x1 + 2x2 = 12 intersect, namely at (x1 , x2 ) = (20/3, 8/3),
because this would violate the second constraint x1 + 4x2 ≤ 16. Instead, as a carefully
drawn graph shows, the solution occurs where the first and second constraints both
bind, at the intersection of the two lines 2x1 + x2 = 16 and x1 + 4x2 = 16, namely
at (x1 , x2 ) = (48/7, 16/7). The resulting profit is 27200/7 = 3885 57 < 3900. So it is
division 1 that should have its capacity increased. Indeed, if the capacity of division 3
is increased by 1 hour per day, some of that increase has to go to waste because of the
limited capacities in divisions 1 and 2.
91
17.4 A general economic interpretation
2. (a) The problem is similar to Exercise 17.3.3. The linear program is set out in the book. As
a carefully drawn graph will show, the solution occurs at the intersection of the two lines
6x1 + 3x2 = 54 and 5x1 + 5x2 = 50, where (x1 , x2 ) = (8, 2). Note that the maximum
profit is 300 · 8 + 200 · 2 = 2800.
(b) The dual problem is
min (54u1 + 48u2 + 50u3 )
s.t.


 6u1 + 4u2 + 5u3 ≥ 300
3u1 + 6u2 + 5u3 ≥ 200


u 1 , u 2 , u3 ≥ 0
The optimal solution of the primal is x∗1 = 8, x∗2 = 2. Since the optimal solution of the
primal has both x∗1 and x∗2 positive, the first two constraints in the dual are satisfied
with equality at the optimal triple (u∗1 , u∗2 , u∗3 ). But the second constraint of the primal
problem is satisfied with inequality, because 4x∗1 + 6x∗2 = 44 < 48. Hence u∗2 = 0, with
6u∗1 + 5u∗3 = 300 and 3u∗1 + 5u∗3 = 200. It follows that u∗1 = 100/3, u∗2 = 0, and u∗3 = 20.
Moreover, the optimum value 54u∗1 + 48u∗2 + 50u∗3 = 2800 of the dual equals the optimum
value of the primal.
(c) See the answer given in the book.
17.5 Complementary slackness
3. (a) See the answer given in the book.
(b) The dual is given in the text. It is illustrated graphically in Fig. SM17.5.3, where (1)
labels the first, (2) the second, and (3) the third constraint. The parallel dashed lines are
level curves of the objective 300x1 +500x2 . We see from the figure that optimum occurs at
the point where the first and the third constraint are satisfied with equality—i.e., where
10x∗1 + 20x∗2 = 10 000 and 20x∗1 + 20x∗2 = 11 000. The solution is x∗1 = 100 and x∗2 = 450.
The maximum value of the objective function is 300 · 100 + 500 · 450 = 255 000.
Since both variables in the dual problem are positive at the optimum, complementary
slackness implies that the two constraints in the primal problem must be satisfied with
equality at the optimum, that is,
10y1∗ + 20y2∗ + 20y3∗ = 300
20y1∗ + 10y2∗ + 20y3∗ = 500
Moreover, the second constraint in the dual problem is satisfied with strict inequality at
the optimum, since 20x∗1 + 10x∗2 = 6500 < 8000. Therefore the corresponding variable
in the primal problem must be zero at the optimum, so y2∗ = 0. Together with the two
equations above this implies that y1∗ = 20, y3∗ = 5. The maximum value of the objective
function is 10 000 · 20 + 8 000 · 0 + 11 000 · 5 = 255 000.
(c) If the cost per hour in factory 1 increases by 100, this has no effect on the constraints
in the primal, but does increase the right-hand side of the first constraint of the dual by
100. An approximate answer of 100 · 20 = 2000 is the increase in cost that would result
92
x2
1000
900
800
(2)
700
600
500
400
(1)
300
200
(3)
100
100 200 300 400 500 600 700 800 900 1000
x1
Figure SM17.5.3
from choosing the same feasible point (y1∗ , y2∗ , y3∗ ) = (20, 0, 5) in the primal. This may
over-estimate the increased minimum cost, however, because it may be better to switch
some production away from factory 1, now it has become more expensive.
To find the exact answer, we check whether any production is switched by considering
the dual linear program again. It is unchanged except that the first constraint becomes
10x1 + 20x2 ≤ 10100. Looking again at Fig. SM17.5.3, the solution to the dual still
occurs at the intersection of the lines (1) and (3), even after (1) has been shifted out.
In particular, therefore, the solution to the altered primal satisfies exactly the same
constraints (including nonnegativity constraints) with equality, and is therefore the same
point. So the earlier estimate of 2000 for the increased cost is indeed accurate.
Review exercises for Chapter 17
2. (a) Regard the given LP problem as the primal and denote it by (P). Its dual is shown in
answer section of the book. Let us denote that by (D). If you draw the feasible set for
(D) and a line −x1 + x2 = c, you see that as c increases, the line moves to the north-west.
The line with the largest value of c which intersects the feasible set does so at the point
(0, 8), at the intersection of three lines −x1 + 2x2 = 16, −2x1 − x2 = −8, and x1 = 0 that
define three of the constraints, but the constraint x1 ≥ 0 is redundant.
(b) We see that when x1 = 0 and x2 = 8, the second and fourth constraints in (D) are satisfied
with strict inequality, so y2 = y4 = 0 at the optimum of (P). Also, since x2 = 8 > 0,
the second constraint in (P) must be satisfied with equality at the optimum — i.e.,
2y1 − y3 = 1. But then we see that the objective function in (P) can be reduced from
16y1 +6y2 −8y3 −15y4 to 16y1 −8y3 = 8(2y1 −y3 ) = 8. We conclude that any (y1 , y2 , y3 , y4 )
of the form (y1 , y2 , y3 , y4 ) = ( 12 (1 + b), 0, b, 0) must solve (P) provided its components are
93
nonnegative and the first constraint in (P) is satisfied.45 The first constraint reduces to
− 21 (1 + b) − 2b ≥ −1, or b ≤ 51 . We conclude that ( 21 (1 + b), 0, b, 0) is optimal provided
0 ≤ b ≤ 15 .
(c) The objective function in (D) changes to kx1 + x2 , but the constraints remain the same.
The solution (0, 8) found in part (a) also remains unchanged provided that the slope −k
of the level curve x2 = 8 − kx1 through the point (0, 8) remains positive and no less than
the slope 1/2 of the line −x1 + 2x2 = 16. Hence k ≤ − 21 .
4. (a) If a = 0, it is a linear programming problem, whose answer appears in the main text.
(b) When a ≥ 0 we follow the techniques in Section 14.10 and consider the Lagrangian
L = (500 − ax1 )x1 + 250x2 − λ1 (0.04x1 + 0.03x2 − 100)
− λ2 (0.025x1 + 0.05x2 − 100) − λ3 (0.05x1 − 100) − λ4 (0.08x2 − 100).
Then the Kuhn–Tucker conditions, with nonnegativity constraints, are that there exist
numbers λ1 , λ2 , λ3 , and λ4 , such that
(i) ∂L/∂x1 = 500 − 2ax1 − 0.04λ1 − 0.025λ2 − 0.05λ3 ≤ 0, with equality if x1 > 0;
(ii) ∂L/∂x2 = 250 − 0.03λ1 − 0.05λ2 − 0.08λ4 ≤ 0, with equality if x2 > 0;
(iii) λ1 ≥ 0, and λ1 = 0 if 0.04x1 + 0.03x2 < 100;
(iv) λ2 ≥ 0, and λ2 = 0 if 0.025x1 + 0.05x2 < 100;
(v) λ3 ≥ 0, and λ3 = 0 if 0.05x1 < 100;
(vi) λ4 ≥ 0, and λ4 = 0 if 0.08x2 < 100.
(c) The Kuhn–Tucker conditions are sufficient for optimality since the Lagrangian is easily
seen to be concave in (x1 , x2 ) for a ≥ 0. To find when (x1 , x2 ) = (2000, 2000/3) remains
optimal, we must find Lagrange multipliers such that all the Kuhn–Tucker conditions are
still satisfied. Note that (i) and (ii) must still be satisfied with equality. Moreover, the
inequalities in (iv) and (vi) are strict, so λ2 = λ4 = 0. Then (ii) gives λ1 = 25000/3.
It remains to check for which values of a one can satisfy (i) for a λ3 that also satisfies
(v). From (i), 0.05λ3 = 500 − 4000a − 0.04(25000/3) = 500/3 − 4000a ≥ 0 if and only if
a ≤ 1/24.
5. (a) See the answer in the main text.
(b) The admissible set is shown as the shaded infinite polygon in Fig. SM17.R.5. Lines I, II,
and III show where the respective constraints are satisfied with equality. The dotted lines
are level curves for the objective function 100y1 +100y2 . We see that the objective function
has its smallest value at P , the intersection of the two lines I and II. The coordinates
(y1∗ , y2∗ ) for P are given by the two equations 3y1∗ + 2y2∗ = 6 and y1∗ + 2y2∗ = 3. Thus P =
(y1∗ , y2∗ ) = (3/2, 3/4). The optimal value of the objective function is 100(y1∗ + y2∗ ) = 225.
(c) Since the dual has a solution, the duality theorem tells us that (a) also has an optimal
solution, which we denote by (x∗1 , x∗2 , x∗3 ). Since the third constraint in the dual is satisfied
with strict inequality, we must have x∗3 = 0. Moreover, both constraints in the dual must
be satisfied with equality at the optimum because both dual variables are positive in
45
The second constraint we already know is satisfied with equality.
94
y2
5
4
3
100y1 + 100y2 = constant
2
1
P
y1
1
III
2
3
I
4
5
6
7
II
Figure SM17.R.5
optimum. Hence 3x∗1 + x∗2 = 100 and 2x∗1 + 2x∗2 = 100, which gives x∗1 = x∗2 = 25. The
maximal profit is 6x∗1 + 3x∗2 + 4x∗3 = 225, equal to the value of the dual, as expected.
(d) To a first-order approximation, profit increases by y1∗ ∆b1 = 1.5, so the new maximal
profit is 226.5. For this approximation to be exact, the optimal point in the dual must
not change when b1 is increased from 100 to 101. This is obviously true, as one sees from
Fig. SM17.R.5.
(e) The maximum value in the primal is equal to the minimum value in the dual. Given the
same approximation as in part (d), this equals b1 y1∗ +b2 y2∗ , which is obviously homogeneous
of degree 1 in b1 and b2 . More generally, let F (b1 , b2 ) denote the minimum value of the
dual set out in part (b).
Given any α > 0, note that minimizing αb1 y1 + αb2 y2 over the constraint set of the dual
gives the same solution (y1∗ , y2∗ ) as minimizing b1 y1 + b2 y2 over the same constraint set.
Hence F (αb1 , αb2 ) = αF (b1 , b2 ) for all α > 0, so F is homogeneous of degree 1.
95
Download